fs/proc/base.c

   1 /*
   2  *  linux/fs/proc/base.c
   3  *
   4  *  Copyright (C) 1991, 1992 Linus Torvalds
   5  *
   6  *  proc base directory handling functions
   7  *
   8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
   9  *  Instead of using magical inumbers to determine the kind of object
  10  *  we allocate and fill in-core inodes upon lookup. They don't even
  11  *  go into icache. We cache the reference to task_struct upon lookup too.
  12  *  Eventually it should become a filesystem in its own. We don't use the
  13  *  rest of procfs anymore.
  14  *
  15  *
  16  *  Changelog:
  17  *  17-Jan-2005
  18  *  Allan Bezerra
  19  *  Bruna Moreira <bruna.moreira@indt.org.br>
  20  *  Edjard Mota <edjard.mota@indt.org.br>
  21  *  Ilias Biris <ilias.biris@indt.org.br>
  22  *  Mauricio Lin <mauricio.lin@indt.org.br>
  23  *
  24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  25  *
  26  *  A new process specific entry (smaps) included in /proc. It shows the
  27  *  size of rss for each memory area. The maps entry lacks information
  28  *  about physical memory size (rss) for each mapped file, i.e.,
  29  *  rss information for executables and library files.
  30  *  This additional information is useful for any tools that need to know
  31  *  about physical memory consumption for a process specific library.
  32  *
  33  *  Changelog:
  34  *  21-Feb-2005
  35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  36  *  Pud inclusion in the page table walking.
  37  *
  38  *  ChangeLog:
  39  *  10-Mar-2005
  40  *  10LE Instituto Nokia de Tecnologia - INdT:
  41  *  A better way to walks through the page table as suggested by Hugh Dickins.
  42  *
  43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
  44  *  Smaps information related to shared, private, clean and dirty pages.
  45  *
  46  *  Paul Mundt <paul.mundt@nokia.com>:
  47  *  Overall revision about smaps.
  48  */
  49
  50 #include <asm/uaccess.h>
  51
  52 #include <linux/errno.h>
  53 #include <linux/time.h>
  54 #include <linux/proc_fs.h>
  55 #include <linux/stat.h>
  56 #include <linux/task_io_accounting_ops.h>
  57 #include <linux/init.h>
  58 #include <linux/capability.h>
  59 #include <linux/file.h>
  60 #include <linux/fdtable.h>
  61 #include <linux/string.h>
  62 #include <linux/seq_file.h>
  63 #include <linux/namei.h>
  64 #include <linux/mnt_namespace.h>
  65 #include <linux/mm.h>
  66 #include <linux/swap.h>
  67 #include <linux/rcupdate.h>
  68 #include <linux/kallsyms.h>
  69 #include <linux/stacktrace.h>
  70 #include <linux/resource.h>
  71 #include <linux/module.h>
  72 #include <linux/mount.h>
  73 #include <linux/security.h>
  74 #include <linux/ptrace.h>
  75 #include <linux/tracehook.h>
  76 #include <linux/cgroup.h>
  77 #include <linux/cpuset.h>
  78 #include <linux/audit.h>
  79 #include <linux/poll.h>
  80 #include <linux/nsproxy.h>
  81 #include <linux/oom.h>
  82 #include <linux/elf.h>
  83 #include <linux/pid_namespace.h>
  84 #include <linux/user_namespace.h>
  85 #include <linux/fs_struct.h>
  86 #include <linux/slab.h>
  87 #include <linux/flex_array.h>
  88 #ifdef CONFIG_HARDWALL
  89 #include <asm/hardwall.h>
  90 #endif
  91 #include <trace/events/oom.h>
  92 #include "internal.h"
  93
  94 /* NOTE:
  95  *      Implementing inode permission operations in /proc is almost
  96  *      certainly an error.  Permission checks need to happen during
  97  *      each system call not at open time.  The reason is that most of
  98  *      what we wish to check for permissions in /proc varies at runtime.
  99  *
 100  *      The classic example of a problem is opening file descriptors
 101  *      in /proc for a task before it execs a suid executable.
 102  */
 103
 104 struct pid_entry {
 105         char *name;
 106         int len;
 107         umode_t mode;
 108         const struct inode_operations *iop;
 109         const struct file_operations *fop;
 110         union proc_op op;
 111 };
 112
 113 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 114         .name = (NAME),                                 \
 115         .len  = sizeof(NAME) - 1,                       \
 116         .mode = MODE,                                   \
 117         .iop  = IOP,                                    \
 118         .fop  = FOP,                                    \
 119         .op   = OP,                                     \
 120 }
 121
 122 #define DIR(NAME, MODE, iops, fops)     \
 123         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 124 #define LNK(NAME, get_link)                                     \
 125         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 126                 &proc_pid_link_inode_operations, NULL,          \
 127                 { .proc_get_link = get_link } )
 128 #define REG(NAME, MODE, fops)                           \
 129         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 130 #define INF(NAME, MODE, read)                           \
 131         NOD(NAME, (S_IFREG|(MODE)),                     \
 132                 NULL, &proc_info_file_operations,       \
 133                 { .proc_read = read } )
 134 #define ONE(NAME, MODE, show)                           \
 135         NOD(NAME, (S_IFREG|(MODE)),                     \
 136                 NULL, &proc_single_file_operations,     \
 137                 { .proc_show = show } )
 138
 139 static int proc_fd_permission(struct inode *inode, int mask);
 140
 141 /*
 142  * Count the number of hardlinks for the pid_entry table, excluding the .
 143  * and .. links.
 144  */
 145 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 146         unsigned int n)
 147 {
 148         unsigned int i;
 149         unsigned int count;
 150
 151         count = 0;
 152         for (i = 0; i < n; ++i) {
 153                 if (S_ISDIR(entries[i].mode))
 154                         ++count;
 155         }
 156
 157         return count;
 158 }
 159
 160 static int get_task_root(struct task_struct *task, struct path *root)
 161 {
 162         int result = -ENOENT;
 163
 164         task_lock(task);
 165         if (task->fs) {
 166                 get_fs_root(task->fs, root);
 167                 result = 0;
 168         }
 169         task_unlock(task);
 170         return result;
 171 }
 172
 173 static int proc_cwd_link(struct dentry *dentry, struct path *path)
 174 {
 175         struct task_struct *task = get_proc_task(dentry->d_inode);
 176         int result = -ENOENT;
 177
 178         if (task) {
 179                 task_lock(task);
 180                 if (task->fs) {
 181                         get_fs_pwd(task->fs, path);
 182                         result = 0;
 183                 }
 184                 task_unlock(task);
 185                 put_task_struct(task);
 186         }
 187         return result;
 188 }
 189
 190 static int proc_root_link(struct dentry *dentry, struct path *path)
 191 {
 192         struct task_struct *task = get_proc_task(dentry->d_inode);
 193         int result = -ENOENT;
 194
 195         if (task) {
 196                 result = get_task_root(task, path);
 197                 put_task_struct(task);
 198         }
 199         return result;
 200 }
 201
 202 struct mm_struct *mm_for_maps(struct task_struct *task)
 203 {
 204         return mm_access(task, PTRACE_MODE_READ);
 205 }
 206
 207 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 208 {
 209         int res = 0;
 210         unsigned int len;
 211         struct mm_struct *mm = get_task_mm(task);
 212         if (!mm)
 213                 goto out;
 214         if (!mm->arg_end)
 215                 goto out_mm;    /* Shh! No looking before we're done */
 216
 217         len = mm->arg_end - mm->arg_start;
 218
 219         if (len > PAGE_SIZE)
 220                 len = PAGE_SIZE;
 221
 222         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 223
 224         // If the nul at the end of args has been overwritten, then
 225         // assume application is using setproctitle(3).
 226         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 227                 len = strnlen(buffer, res);
 228                 if (len < res) {
 229                     res = len;
 230                 } else {
 231                         len = mm->env_end - mm->env_start;
 232                         if (len > PAGE_SIZE - res)
 233                                 len = PAGE_SIZE - res;
 234                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 235                         res = strnlen(buffer, res);
 236                 }
 237         }
 238 out_mm:
 239         mmput(mm);
 240 out:
 241         return res;
 242 }
 243
 244 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 245 {
 246         struct mm_struct *mm = mm_for_maps(task);
 247         int res = PTR_ERR(mm);
 248         if (mm && !IS_ERR(mm)) {
 249                 unsigned int nwords = 0;
 250                 do {
 251                         nwords += 2;
 252                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 253                 res = nwords * sizeof(mm->saved_auxv[0]);
 254                 if (res > PAGE_SIZE)
 255                         res = PAGE_SIZE;
 256                 memcpy(buffer, mm->saved_auxv, res);
 257                 mmput(mm);
 258         }
 259         return res;
 260 }
 261
 262
 263 #ifdef CONFIG_KALLSYMS
 264 /*
 265  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 266  * Returns the resolved symbol.  If that fails, simply return the address.
 267  */
 268 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 269 {
 270         unsigned long wchan;
 271         char symname[KSYM_NAME_LEN];
 272
 273         wchan = get_wchan(task);
 274
 275         if (lookup_symbol_name(wchan, symname) < 0)
 276                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
 277                         return 0;
 278                 else
 279                         return sprintf(buffer, "%lu", wchan);
 280         else
 281                 return sprintf(buffer, "%s", symname);
 282 }
 283 #endif /* CONFIG_KALLSYMS */
 284
 285 static int lock_trace(struct task_struct *task)
 286 {
 287         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 288         if (err)
 289                 return err;
 290         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
 291                 mutex_unlock(&task->signal->cred_guard_mutex);
 292                 return -EPERM;
 293         }
 294         return 0;
 295 }
 296
 297 static void unlock_trace(struct task_struct *task)
 298 {
 299         mutex_unlock(&task->signal->cred_guard_mutex);
 300 }
 301
 302 #ifdef CONFIG_STACKTRACE
 303
 304 #define MAX_STACK_TRACE_DEPTH   64
 305
 306 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 307                           struct pid *pid, struct task_struct *task)
 308 {
 309         struct stack_trace trace;
 310         unsigned long *entries;
 311         int err;
 312         int i;
 313
 314         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 315         if (!entries)
 316                 return -ENOMEM;
 317
 318         trace.nr_entries        = 0;
 319         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 320         trace.entries           = entries;
 321         trace.skip              = 0;
 322
 323         err = lock_trace(task);
 324         if (!err) {
 325                 save_stack_trace_tsk(task, &trace);
 326
 327                 for (i = 0; i < trace.nr_entries; i++) {
 328                         seq_printf(m, "[<%pK>] %pS\n",
 329                                    (void *)entries[i], (void *)entries[i]);
 330                 }
 331                 unlock_trace(task);
 332         }
 333         kfree(entries);
 334
 335         return err;
 336 }
 337 #endif
 338
 339 #ifdef CONFIG_SCHEDSTATS
 340 /*
 341  * Provides /proc/PID/schedstat
 342  */
 343 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 344 {
 345         return sprintf(buffer, "%llu %llu %lu\n",
 346                         (unsigned long long)task->se.sum_exec_runtime,
 347                         (unsigned long long)task->sched_info.run_delay,
 348                         task->sched_info.pcount);
 349 }
 350 #endif
 351
 352 #ifdef CONFIG_LATENCYTOP
 353 static int lstats_show_proc(struct seq_file *m, void *v)
 354 {
 355         int i;
 356         struct inode *inode = m->private;
 357         struct task_struct *task = get_proc_task(inode);
 358
 359         if (!task)
 360                 return -ESRCH;
 361         seq_puts(m, "Latency Top version : v0.1\n");
 362         for (i = 0; i < 32; i++) {
 363                 struct latency_record *lr = &task->latency_record[i];
 364                 if (lr->backtrace[0]) {
 365                         int q;
 366                         seq_printf(m, "%i %li %li",
 367                                    lr->count, lr->time, lr->max);
 368                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 369                                 unsigned long bt = lr->backtrace[q];
 370                                 if (!bt)
 371                                         break;
 372                                 if (bt == ULONG_MAX)
 373                                         break;
 374                                 seq_printf(m, " %ps", (void *)bt);
 375                         }
 376                         seq_putc(m, '\n');
 377                 }
 378
 379         }
 380         put_task_struct(task);
 381         return 0;
 382 }
 383
 384 static int lstats_open(struct inode *inode, struct file *file)
 385 {
 386         return single_open(file, lstats_show_proc, inode);
 387 }
 388
 389 static ssize_t lstats_write(struct file *file, const char __user *buf,
 390                             size_t count, loff_t *offs)
 391 {
 392         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 393
 394         if (!task)
 395                 return -ESRCH;
 396         clear_all_latency_tracing(task);
 397         put_task_struct(task);
 398
 399         return count;
 400 }
 401
 402 static const struct file_operations proc_lstats_operations = {
 403         .open           = lstats_open,
 404         .read           = seq_read,
 405         .write          = lstats_write,
 406         .llseek         = seq_lseek,
 407         .release        = single_release,
 408 };
 409
 410 #endif
 411
 412 static int proc_oom_score(struct task_struct *task, char *buffer)
 413 {
 414         unsigned long totalpages = totalram_pages + total_swap_pages;
 415         unsigned long points = 0;
 416
 417         read_lock(&tasklist_lock);
 418         if (pid_alive(task))
 419                 points = oom_badness(task, NULL, NULL, totalpages) *
 420                                                 1000 / totalpages;
 421         read_unlock(&tasklist_lock);
 422         return sprintf(buffer, "%lu\n", points);
 423 }
 424
 425 struct limit_names {
 426         char *name;
 427         char *unit;
 428 };
 429
 430 static const struct limit_names lnames[RLIM_NLIMITS] = {
 431         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 432         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 433         [RLIMIT_DATA] = {"Max data size", "bytes"},
 434         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 435         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 436         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 437         [RLIMIT_NPROC] = {"Max processes", "processes"},
 438         [RLIMIT_NOFILE] = {"Max open files", "files"},
 439         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 440         [RLIMIT_AS] = {"Max address space", "bytes"},
 441         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 442         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 443         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 444         [RLIMIT_NICE] = {"Max nice priority", NULL},
 445         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 446         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 447 };
 448
 449 /* Display limits for a process */
 450 static int proc_pid_limits(struct task_struct *task, char *buffer)
 451 {
 452         unsigned int i;
 453         int count = 0;
 454         unsigned long flags;
 455         char *bufptr = buffer;
 456
 457         struct rlimit rlim[RLIM_NLIMITS];
 458
 459         if (!lock_task_sighand(task, &flags))
 460                 return 0;
 461         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 462         unlock_task_sighand(task, &flags);
 463
 464         /*
 465          * print the file header
 466          */
 467         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 468                         "Limit", "Soft Limit", "Hard Limit", "Units");
 469
 470         for (i = 0; i < RLIM_NLIMITS; i++) {
 471                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 472                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 473                                          lnames[i].name, "unlimited");
 474                 else
 475                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 476                                          lnames[i].name, rlim[i].rlim_cur);
 477
 478                 if (rlim[i].rlim_max == RLIM_INFINITY)
 479                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 480                 else
 481                         count += sprintf(&bufptr[count], "%-20lu ",
 482                                          rlim[i].rlim_max);
 483
 484                 if (lnames[i].unit)
 485                         count += sprintf(&bufptr[count], "%-10s\n",
 486                                          lnames[i].unit);
 487                 else
 488                         count += sprintf(&bufptr[count], "\n");
 489         }
 490
 491         return count;
 492 }
 493
 494 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 495 static int proc_pid_syscall(struct task_struct *task, char *buffer)
 496 {
 497         long nr;
 498         unsigned long args[6], sp, pc;
 499         int res = lock_trace(task);
 500         if (res)
 501                 return res;
 502
 503         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 504                 res = sprintf(buffer, "running\n");
 505         else if (nr < 0)
 506                 res = sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 507         else
 508                 res = sprintf(buffer,
 509                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 510                        nr,
 511                        args[0], args[1], args[2], args[3], args[4], args[5],
 512                        sp, pc);
 513         unlock_trace(task);
 514         return res;
 515 }
 516 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 517
 518 /************************************************************************/
 519 /*                       Here the fs part begins                        */
 520 /************************************************************************/
 521
 522 /* permission checks */
 523 static int proc_fd_access_allowed(struct inode *inode)
 524 {
 525         struct task_struct *task;
 526         int allowed = 0;
 527         /* Allow access to a task's file descriptors if it is us or we
 528          * may use ptrace attach to the process and find out that
 529          * information.
 530          */
 531         task = get_proc_task(inode);
 532         if (task) {
 533                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 534                 put_task_struct(task);
 535         }
 536         return allowed;
 537 }
 538
 539 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 540 {
 541         int error;
 542         struct inode *inode = dentry->d_inode;
 543
 544         if (attr->ia_valid & ATTR_MODE)
 545                 return -EPERM;
 546
 547         error = inode_change_ok(inode, attr);
 548         if (error)
 549                 return error;
 550
 551         if ((attr->ia_valid & ATTR_SIZE) &&
 552             attr->ia_size != i_size_read(inode)) {
 553                 error = vmtruncate(inode, attr->ia_size);
 554                 if (error)
 555                         return error;
 556         }
 557
 558         setattr_copy(inode, attr);
 559         mark_inode_dirty(inode);
 560         return 0;
 561 }
 562
 563 /*
 564  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 565  * or euid/egid (for hide_pid_min=2)?
 566  */
 567 static bool has_pid_permissions(struct pid_namespace *pid,
 568                                  struct task_struct *task,
 569                                  int hide_pid_min)
 570 {
 571         if (pid->hide_pid < hide_pid_min)
 572                 return true;
 573         if (in_group_p(pid->pid_gid))
 574                 return true;
 575         return ptrace_may_access(task, PTRACE_MODE_READ);
 576 }
 577
 578
 579 static int proc_pid_permission(struct inode *inode, int mask)
 580 {
 581         struct pid_namespace *pid = inode->i_sb->s_fs_info;
 582         struct task_struct *task;
 583         bool has_perms;
 584
 585         task = get_proc_task(inode);
 586         if (!task)
 587                 return -ESRCH;
 588         has_perms = has_pid_permissions(pid, task, 1);
 589         put_task_struct(task);
 590
 591         if (!has_perms) {
 592                 if (pid->hide_pid == 2) {
 593                         /*
 594                          * Let's make getdents(), stat(), and open()
 595                          * consistent with each other.  If a process
 596                          * may not stat() a file, it shouldn't be seen
 597                          * in procfs at all.
 598                          */
 599                         return -ENOENT;
 600                 }
 601
 602                 return -EPERM;
 603         }
 604         return generic_permission(inode, mask);
 605 }
 606
 607
 608
 609 static const struct inode_operations proc_def_inode_operations = {
 610         .setattr        = proc_setattr,
 611 };
 612
 613 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 614
 615 static ssize_t proc_info_read(struct file * file, char __user * buf,
 616                           size_t count, loff_t *ppos)
 617 {
 618         struct inode * inode = file->f_path.dentry->d_inode;
 619         unsigned long page;
 620         ssize_t length;
 621         struct task_struct *task = get_proc_task(inode);
 622
 623         length = -ESRCH;
 624         if (!task)
 625                 goto out_no_task;
 626
 627         if (count > PROC_BLOCK_SIZE)
 628                 count = PROC_BLOCK_SIZE;
 629
 630         length = -ENOMEM;
 631         if (!(page = __get_free_page(GFP_TEMPORARY)))
 632                 goto out;
 633
 634         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 635
 636         if (length >= 0)
 637                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 638         free_page(page);
 639 out:
 640         put_task_struct(task);
 641 out_no_task:
 642         return length;
 643 }
 644
 645 static const struct file_operations proc_info_file_operations = {
 646         .read           = proc_info_read,
 647         .llseek         = generic_file_llseek,
 648 };
 649
 650 static int proc_single_show(struct seq_file *m, void *v)
 651 {
 652         struct inode *inode = m->private;
 653         struct pid_namespace *ns;
 654         struct pid *pid;
 655         struct task_struct *task;
 656         int ret;
 657
 658         ns = inode->i_sb->s_fs_info;
 659         pid = proc_pid(inode);
 660         task = get_pid_task(pid, PIDTYPE_PID);
 661         if (!task)
 662                 return -ESRCH;
 663
 664         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 665
 666         put_task_struct(task);
 667         return ret;
 668 }
 669
 670 static int proc_single_open(struct inode *inode, struct file *filp)
 671 {
 672         return single_open(filp, proc_single_show, inode);
 673 }
 674
 675 static const struct file_operations proc_single_file_operations = {
 676         .open           = proc_single_open,
 677         .read           = seq_read,
 678         .llseek         = seq_lseek,
 679         .release        = single_release,
 680 };
 681
 682 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 683 {
 684         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 685         struct mm_struct *mm;
 686
 687         if (!task)
 688                 return -ESRCH;
 689
 690         mm = mm_access(task, mode);
 691         put_task_struct(task);
 692
 693         if (IS_ERR(mm))
 694                 return PTR_ERR(mm);
 695
 696         if (mm) {
 697                 /* ensure this mm_struct can't be freed */
 698                 atomic_inc(&mm->mm_count);
 699                 /* but do not pin its memory */
 700                 mmput(mm);
 701         }
 702
 703         /* OK to pass negative loff_t, we can catch out-of-range */
 704         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 705         file->private_data = mm;
 706
 707         return 0;
 708 }
 709
 710 static int mem_open(struct inode *inode, struct file *file)
 711 {
 712         return __mem_open(inode, file, PTRACE_MODE_ATTACH);
 713 }
 714
 715 static ssize_t mem_rw(struct file *file, char __user *buf,
 716                         size_t count, loff_t *ppos, int write)
 717 {
 718         struct mm_struct *mm = file->private_data;
 719         unsigned long addr = *ppos;
 720         ssize_t copied;
 721         char *page;
 722
 723         if (!mm)
 724                 return 0;
 725
 726         page = (char *)__get_free_page(GFP_TEMPORARY);
 727         if (!page)
 728                 return -ENOMEM;
 729
 730         copied = 0;
 731         if (!atomic_inc_not_zero(&mm->mm_users))
 732                 goto free;
 733
 734         while (count > 0) {
 735                 int this_len = min_t(int, count, PAGE_SIZE);
 736
 737                 if (write && copy_from_user(page, buf, this_len)) {
 738                         copied = -EFAULT;
 739                         break;
 740                 }
 741
 742                 this_len = access_remote_vm(mm, addr, page, this_len, write);
 743                 if (!this_len) {
 744                         if (!copied)
 745                                 copied = -EIO;
 746                         break;
 747                 }
 748
 749                 if (!write && copy_to_user(buf, page, this_len)) {
 750                         copied = -EFAULT;
 751                         break;
 752                 }
 753
 754                 buf += this_len;
 755                 addr += this_len;
 756                 copied += this_len;
 757                 count -= this_len;
 758         }
 759         *ppos = addr;
 760
 761         mmput(mm);
 762 free:
 763         free_page((unsigned long) page);
 764         return copied;
 765 }
 766
 767 static ssize_t mem_read(struct file *file, char __user *buf,
 768                         size_t count, loff_t *ppos)
 769 {
 770         return mem_rw(file, buf, count, ppos, 0);
 771 }
 772
 773 static ssize_t mem_write(struct file *file, const char __user *buf,
 774                          size_t count, loff_t *ppos)
 775 {
 776         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 777 }
 778
 779 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 780 {
 781         switch (orig) {
 782         case 0:
 783                 file->f_pos = offset;
 784                 break;
 785         case 1:
 786                 file->f_pos += offset;
 787                 break;
 788         default:
 789                 return -EINVAL;
 790         }
 791         force_successful_syscall_return();
 792         return file->f_pos;
 793 }
 794
 795 static int mem_release(struct inode *inode, struct file *file)
 796 {
 797         struct mm_struct *mm = file->private_data;
 798         if (mm)
 799                 mmdrop(mm);
 800         return 0;
 801 }
 802
 803 static const struct file_operations proc_mem_operations = {
 804         .llseek         = mem_lseek,
 805         .read           = mem_read,
 806         .write          = mem_write,
 807         .open           = mem_open,
 808         .release        = mem_release,
 809 };
 810
 811 static int environ_open(struct inode *inode, struct file *file)
 812 {
 813         return __mem_open(inode, file, PTRACE_MODE_READ);
 814 }
 815
 816 static ssize_t environ_read(struct file *file, char __user *buf,
 817                         size_t count, loff_t *ppos)
 818 {
 819         char *page;
 820         unsigned long src = *ppos;
 821         int ret = 0;
 822         struct mm_struct *mm = file->private_data;
 823
 824         if (!mm)
 825                 return 0;
 826
 827         page = (char *)__get_free_page(GFP_TEMPORARY);
 828         if (!page)
 829                 return -ENOMEM;
 830
 831         ret = 0;
 832         if (!atomic_inc_not_zero(&mm->mm_users))
 833                 goto free;
 834         while (count > 0) {
 835                 int this_len, retval, max_len;
 836
 837                 this_len = mm->env_end - (mm->env_start + src);
 838
 839                 if (this_len <= 0)
 840                         break;
 841
 842                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 843                 this_len = (this_len > max_len) ? max_len : this_len;
 844
 845                 retval = access_remote_vm(mm, (mm->env_start + src),
 846                         page, this_len, 0);
 847
 848                 if (retval <= 0) {
 849                         ret = retval;
 850                         break;
 851                 }
 852
 853                 if (copy_to_user(buf, page, retval)) {
 854                         ret = -EFAULT;
 855                         break;
 856                 }
 857
 858                 ret += retval;
 859                 src += retval;
 860                 buf += retval;
 861                 count -= retval;
 862         }
 863         *ppos = src;
 864         mmput(mm);
 865
 866 free:
 867         free_page((unsigned long) page);
 868         return ret;
 869 }
 870
 871 static const struct file_operations proc_environ_operations = {
 872         .open           = environ_open,
 873         .read           = environ_read,
 874         .llseek         = generic_file_llseek,
 875         .release        = mem_release,
 876 };
 877
 878 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 879                                 size_t count, loff_t *ppos)
 880 {
 881         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 882         char buffer[PROC_NUMBUF];
 883         size_t len;
 884         int oom_adjust = OOM_DISABLE;
 885         unsigned long flags;
 886
 887         if (!task)
 888                 return -ESRCH;
 889
 890         if (lock_task_sighand(task, &flags)) {
 891                 oom_adjust = task->signal->oom_adj;
 892                 unlock_task_sighand(task, &flags);
 893         }
 894
 895         put_task_struct(task);
 896
 897         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
 898
 899         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 900 }
 901
 902 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
 903                                 size_t count, loff_t *ppos)
 904 {
 905         struct task_struct *task;
 906         char buffer[PROC_NUMBUF];
 907         int oom_adjust;
 908         unsigned long flags;
 909         int err;
 910
 911         memset(buffer, 0, sizeof(buffer));
 912         if (count > sizeof(buffer) - 1)
 913                 count = sizeof(buffer) - 1;
 914         if (copy_from_user(buffer, buf, count)) {
 915                 err = -EFAULT;
 916                 goto out;
 917         }
 918
 919         err = kstrtoint(strstrip(buffer), 0, &oom_adjust);
 920         if (err)
 921                 goto out;
 922         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
 923              oom_adjust != OOM_DISABLE) {
 924                 err = -EINVAL;
 925                 goto out;
 926         }
 927
 928         task = get_proc_task(file->f_path.dentry->d_inode);
 929         if (!task) {
 930                 err = -ESRCH;
 931                 goto out;
 932         }
 933
 934         task_lock(task);
 935         if (!task->mm) {
 936                 err = -EINVAL;
 937                 goto err_task_lock;
 938         }
 939
 940         if (!lock_task_sighand(task, &flags)) {
 941                 err = -ESRCH;
 942                 goto err_task_lock;
 943         }
 944
 945         if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
 946                 err = -EACCES;
 947                 goto err_sighand;
 948         }
 949
 950         /*
 951          * Warn that /proc/pid/oom_adj is deprecated, see
 952          * Documentation/feature-removal-schedule.txt.
 953          */
 954         printk_once(KERN_WARNING "%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
 955                   current->comm, task_pid_nr(current), task_pid_nr(task),
 956                   task_pid_nr(task));
 957         task->signal->oom_adj = oom_adjust;
 958         /*
 959          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
 960          * value is always attainable.
 961          */
 962         if (task->signal->oom_adj == OOM_ADJUST_MAX)
 963                 task->signal->oom_score_adj = OOM_SCORE_ADJ_MAX;
 964         else
 965                 task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
 966                                                                 -OOM_DISABLE;
 967         trace_oom_score_adj_update(task);
 968 err_sighand:
 969         unlock_task_sighand(task, &flags);
 970 err_task_lock:
 971         task_unlock(task);
 972         put_task_struct(task);
 973 out:
 974         return err < 0 ? err : count;
 975 }
 976
 977 static const struct file_operations proc_oom_adjust_operations = {
 978         .read           = oom_adjust_read,
 979         .write          = oom_adjust_write,
 980         .llseek         = generic_file_llseek,
 981 };
 982
 983 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
 984                                         size_t count, loff_t *ppos)
 985 {
 986         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 987         char buffer[PROC_NUMBUF];
 988         int oom_score_adj = OOM_SCORE_ADJ_MIN;
 989         unsigned long flags;
 990         size_t len;
 991
 992         if (!task)
 993                 return -ESRCH;
 994         if (lock_task_sighand(task, &flags)) {
 995                 oom_score_adj = task->signal->oom_score_adj;
 996                 unlock_task_sighand(task, &flags);
 997         }
 998         put_task_struct(task);
 999         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj);
1000         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1001 }
1002
1003 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1004                                         size_t count, loff_t *ppos)
1005 {
1006         struct task_struct *task;
1007         char buffer[PROC_NUMBUF];
1008         unsigned long flags;
1009         int oom_score_adj;
1010         int err;
1011
1012         memset(buffer, 0, sizeof(buffer));
1013         if (count > sizeof(buffer) - 1)
1014                 count = sizeof(buffer) - 1;
1015         if (copy_from_user(buffer, buf, count)) {
1016                 err = -EFAULT;
1017                 goto out;
1018         }
1019
1020         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1021         if (err)
1022                 goto out;
1023         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1024                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1025                 err = -EINVAL;
1026                 goto out;
1027         }
1028
1029         task = get_proc_task(file->f_path.dentry->d_inode);
1030         if (!task) {
1031                 err = -ESRCH;
1032                 goto out;
1033         }
1034
1035         task_lock(task);
1036         if (!task->mm) {
1037                 err = -EINVAL;
1038                 goto err_task_lock;
1039         }
1040
1041         if (!lock_task_sighand(task, &flags)) {
1042                 err = -ESRCH;
1043                 goto err_task_lock;
1044         }
1045
1046         if (oom_score_adj < task->signal->oom_score_adj_min &&
1047                         !capable(CAP_SYS_RESOURCE)) {
1048                 err = -EACCES;
1049                 goto err_sighand;
1050         }
1051
1052         task->signal->oom_score_adj = oom_score_adj;
1053         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1054                 task->signal->oom_score_adj_min = oom_score_adj;
1055         trace_oom_score_adj_update(task);
1056         /*
1057          * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
1058          * always attainable.
1059          */
1060         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1061                 task->signal->oom_adj = OOM_DISABLE;
1062         else
1063                 task->signal->oom_adj = (oom_score_adj * OOM_ADJUST_MAX) /
1064                                                         OOM_SCORE_ADJ_MAX;
1065 err_sighand:
1066         unlock_task_sighand(task, &flags);
1067 err_task_lock:
1068         task_unlock(task);
1069         put_task_struct(task);
1070 out:
1071         return err < 0 ? err : count;
1072 }
1073
1074 static const struct file_operations proc_oom_score_adj_operations = {
1075         .read           = oom_score_adj_read,
1076         .write          = oom_score_adj_write,
1077         .llseek         = default_llseek,
1078 };
1079
1080 #ifdef CONFIG_AUDITSYSCALL
1081 #define TMPBUFLEN 21
1082 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1083                                   size_t count, loff_t *ppos)
1084 {
1085         struct inode * inode = file->f_path.dentry->d_inode;
1086         struct task_struct *task = get_proc_task(inode);
1087         ssize_t length;
1088         char tmpbuf[TMPBUFLEN];
1089
1090         if (!task)
1091                 return -ESRCH;
1092         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1093                                 audit_get_loginuid(task));
1094         put_task_struct(task);
1095         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1096 }
1097
1098 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1099                                    size_t count, loff_t *ppos)
1100 {
1101         struct inode * inode = file->f_path.dentry->d_inode;
1102         char *page, *tmp;
1103         ssize_t length;
1104         uid_t loginuid;
1105
1106         rcu_read_lock();
1107         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1108                 rcu_read_unlock();
1109                 return -EPERM;
1110         }
1111         rcu_read_unlock();
1112
1113         if (count >= PAGE_SIZE)
1114                 count = PAGE_SIZE - 1;
1115
1116         if (*ppos != 0) {
1117                 /* No partial writes. */
1118                 return -EINVAL;
1119         }
1120         page = (char*)__get_free_page(GFP_TEMPORARY);
1121         if (!page)
1122                 return -ENOMEM;
1123         length = -EFAULT;
1124         if (copy_from_user(page, buf, count))
1125                 goto out_free_page;
1126
1127         page[count] = '\0';
1128         loginuid = simple_strtoul(page, &tmp, 10);
1129         if (tmp == page) {
1130                 length = -EINVAL;
1131                 goto out_free_page;
1132
1133         }
1134         length = audit_set_loginuid(loginuid);
1135         if (likely(length == 0))
1136                 length = count;
1137
1138 out_free_page:
1139         free_page((unsigned long) page);
1140         return length;
1141 }
1142
1143 static const struct file_operations proc_loginuid_operations = {
1144         .read           = proc_loginuid_read,
1145         .write          = proc_loginuid_write,
1146         .llseek         = generic_file_llseek,
1147 };
1148
1149 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1150                                   size_t count, loff_t *ppos)
1151 {
1152         struct inode * inode = file->f_path.dentry->d_inode;
1153         struct task_struct *task = get_proc_task(inode);
1154         ssize_t length;
1155         char tmpbuf[TMPBUFLEN];
1156
1157         if (!task)
1158                 return -ESRCH;
1159         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1160                                 audit_get_sessionid(task));
1161         put_task_struct(task);
1162         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1163 }
1164
1165 static const struct file_operations proc_sessionid_operations = {
1166         .read           = proc_sessionid_read,
1167         .llseek         = generic_file_llseek,
1168 };
1169 #endif
1170
1171 #ifdef CONFIG_FAULT_INJECTION
1172 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1173                                       size_t count, loff_t *ppos)
1174 {
1175         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1176         char buffer[PROC_NUMBUF];
1177         size_t len;
1178         int make_it_fail;
1179
1180         if (!task)
1181                 return -ESRCH;
1182         make_it_fail = task->make_it_fail;
1183         put_task_struct(task);
1184
1185         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1186
1187         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1188 }
1189
1190 static ssize_t proc_fault_inject_write(struct file * file,
1191                         const char __user * buf, size_t count, loff_t *ppos)
1192 {
1193         struct task_struct *task;
1194         char buffer[PROC_NUMBUF], *end;
1195         int make_it_fail;
1196
1197         if (!capable(CAP_SYS_RESOURCE))
1198                 return -EPERM;
1199         memset(buffer, 0, sizeof(buffer));
1200         if (count > sizeof(buffer) - 1)
1201                 count = sizeof(buffer) - 1;
1202         if (copy_from_user(buffer, buf, count))
1203                 return -EFAULT;
1204         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1205         if (*end)
1206                 return -EINVAL;
1207         task = get_proc_task(file->f_dentry->d_inode);
1208         if (!task)
1209                 return -ESRCH;
1210         task->make_it_fail = make_it_fail;
1211         put_task_struct(task);
1212
1213         return count;
1214 }
1215
1216 static const struct file_operations proc_fault_inject_operations = {
1217         .read           = proc_fault_inject_read,
1218         .write          = proc_fault_inject_write,
1219         .llseek         = generic_file_llseek,
1220 };
1221 #endif
1222
1223
1224 #ifdef CONFIG_SCHED_DEBUG
1225 /*
1226  * Print out various scheduling related per-task fields:
1227  */
1228 static int sched_show(struct seq_file *m, void *v)
1229 {
1230         struct inode *inode = m->private;
1231         struct task_struct *p;
1232
1233         p = get_proc_task(inode);
1234         if (!p)
1235                 return -ESRCH;
1236         proc_sched_show_task(p, m);
1237
1238         put_task_struct(p);
1239
1240         return 0;
1241 }
1242
1243 static ssize_t
1244 sched_write(struct file *file, const char __user *buf,
1245             size_t count, loff_t *offset)
1246 {
1247         struct inode *inode = file->f_path.dentry->d_inode;
1248         struct task_struct *p;
1249
1250         p = get_proc_task(inode);
1251         if (!p)
1252                 return -ESRCH;
1253         proc_sched_set_task(p);
1254
1255         put_task_struct(p);
1256
1257         return count;
1258 }
1259
1260 static int sched_open(struct inode *inode, struct file *filp)
1261 {
1262         return single_open(filp, sched_show, inode);
1263 }
1264
1265 static const struct file_operations proc_pid_sched_operations = {
1266         .open           = sched_open,
1267         .read           = seq_read,
1268         .write          = sched_write,
1269         .llseek         = seq_lseek,
1270         .release        = single_release,
1271 };
1272
1273 #endif
1274
1275 #ifdef CONFIG_SCHED_AUTOGROUP
1276 /*
1277  * Print out autogroup related information:
1278  */
1279 static int sched_autogroup_show(struct seq_file *m, void *v)
1280 {
1281         struct inode *inode = m->private;
1282         struct task_struct *p;
1283
1284         p = get_proc_task(inode);
1285         if (!p)
1286                 return -ESRCH;
1287         proc_sched_autogroup_show_task(p, m);
1288
1289         put_task_struct(p);
1290
1291         return 0;
1292 }
1293
1294 static ssize_t
1295 sched_autogroup_write(struct file *file, const char __user *buf,
1296             size_t count, loff_t *offset)
1297 {
1298         struct inode *inode = file->f_path.dentry->d_inode;
1299         struct task_struct *p;
1300         char buffer[PROC_NUMBUF];
1301         int nice;
1302         int err;
1303
1304         memset(buffer, 0, sizeof(buffer));
1305         if (count > sizeof(buffer) - 1)
1306                 count = sizeof(buffer) - 1;
1307         if (copy_from_user(buffer, buf, count))
1308                 return -EFAULT;
1309
1310         err = kstrtoint(strstrip(buffer), 0, &nice);
1311         if (err < 0)
1312                 return err;
1313
1314         p = get_proc_task(inode);
1315         if (!p)
1316                 return -ESRCH;
1317
1318         err = proc_sched_autogroup_set_nice(p, nice);
1319         if (err)
1320                 count = err;
1321
1322         put_task_struct(p);
1323
1324         return count;
1325 }
1326
1327 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1328 {
1329         int ret;
1330
1331         ret = single_open(filp, sched_autogroup_show, NULL);
1332         if (!ret) {
1333                 struct seq_file *m = filp->private_data;
1334
1335                 m->private = inode;
1336         }
1337         return ret;
1338 }
1339
1340 static const struct file_operations proc_pid_sched_autogroup_operations = {
1341         .open           = sched_autogroup_open,
1342         .read           = seq_read,
1343         .write          = sched_autogroup_write,
1344         .llseek         = seq_lseek,
1345         .release        = single_release,
1346 };
1347
1348 #endif /* CONFIG_SCHED_AUTOGROUP */
1349
1350 static ssize_t comm_write(struct file *file, const char __user *buf,
1351                                 size_t count, loff_t *offset)
1352 {
1353         struct inode *inode = file->f_path.dentry->d_inode;
1354         struct task_struct *p;
1355         char buffer[TASK_COMM_LEN];
1356
1357         memset(buffer, 0, sizeof(buffer));
1358         if (count > sizeof(buffer) - 1)
1359                 count = sizeof(buffer) - 1;
1360         if (copy_from_user(buffer, buf, count))
1361                 return -EFAULT;
1362
1363         p = get_proc_task(inode);
1364         if (!p)
1365                 return -ESRCH;
1366
1367         if (same_thread_group(current, p))
1368                 set_task_comm(p, buffer);
1369         else
1370                 count = -EINVAL;
1371
1372         put_task_struct(p);
1373
1374         return count;
1375 }
1376
1377 static int comm_show(struct seq_file *m, void *v)
1378 {
1379         struct inode *inode = m->private;
1380         struct task_struct *p;
1381
1382         p = get_proc_task(inode);
1383         if (!p)
1384                 return -ESRCH;
1385
1386         task_lock(p);
1387         seq_printf(m, "%s\n", p->comm);
1388         task_unlock(p);
1389
1390         put_task_struct(p);
1391
1392         return 0;
1393 }
1394
1395 static int comm_open(struct inode *inode, struct file *filp)
1396 {
1397         return single_open(filp, comm_show, inode);
1398 }
1399
1400 static const struct file_operations proc_pid_set_comm_operations = {
1401         .open           = comm_open,
1402         .read           = seq_read,
1403         .write          = comm_write,
1404         .llseek         = seq_lseek,
1405         .release        = single_release,
1406 };
1407
1408 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1409 {
1410         struct task_struct *task;
1411         struct mm_struct *mm;
1412         struct file *exe_file;
1413
1414         task = get_proc_task(dentry->d_inode);
1415         if (!task)
1416                 return -ENOENT;
1417         mm = get_task_mm(task);
1418         put_task_struct(task);
1419         if (!mm)
1420                 return -ENOENT;
1421         exe_file = get_mm_exe_file(mm);
1422         mmput(mm);
1423         if (exe_file) {
1424                 *exe_path = exe_file->f_path;
1425                 path_get(&exe_file->f_path);
1426                 fput(exe_file);
1427                 return 0;
1428         } else
1429                 return -ENOENT;
1430 }
1431
1432 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1433 {
1434         struct inode *inode = dentry->d_inode;
1435         int error = -EACCES;
1436
1437         /* We don't need a base pointer in the /proc filesystem */
1438         path_put(&nd->path);
1439
1440         /* Are we allowed to snoop on the tasks file descriptors? */
1441         if (!proc_fd_access_allowed(inode))
1442                 goto out;
1443
1444         error = PROC_I(inode)->op.proc_get_link(dentry, &nd->path);
1445 out:
1446         return ERR_PTR(error);
1447 }
1448
1449 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1450 {
1451         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1452         char *pathname;
1453         int len;
1454
1455         if (!tmp)
1456                 return -ENOMEM;
1457
1458         pathname = d_path(path, tmp, PAGE_SIZE);
1459         len = PTR_ERR(pathname);
1460         if (IS_ERR(pathname))
1461                 goto out;
1462         len = tmp + PAGE_SIZE - 1 - pathname;
1463
1464         if (len > buflen)
1465                 len = buflen;
1466         if (copy_to_user(buffer, pathname, len))
1467                 len = -EFAULT;
1468  out:
1469         free_page((unsigned long)tmp);
1470         return len;
1471 }
1472
1473 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1474 {
1475         int error = -EACCES;
1476         struct inode *inode = dentry->d_inode;
1477         struct path path;
1478
1479         /* Are we allowed to snoop on the tasks file descriptors? */
1480         if (!proc_fd_access_allowed(inode))
1481                 goto out;
1482
1483         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1484         if (error)
1485                 goto out;
1486
1487         error = do_proc_readlink(&path, buffer, buflen);
1488         path_put(&path);
1489 out:
1490         return error;
1491 }
1492
1493 static const struct inode_operations proc_pid_link_inode_operations = {
1494         .readlink       = proc_pid_readlink,
1495         .follow_link    = proc_pid_follow_link,
1496         .setattr        = proc_setattr,
1497 };
1498
1499
1500 /* building an inode */
1501
1502 static int task_dumpable(struct task_struct *task)
1503 {
1504         int dumpable = 0;
1505         struct mm_struct *mm;
1506
1507         task_lock(task);
1508         mm = task->mm;
1509         if (mm)
1510                 dumpable = get_dumpable(mm);
1511         task_unlock(task);
1512         if(dumpable == 1)
1513                 return 1;
1514         return 0;
1515 }
1516
1517 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1518 {
1519         struct inode * inode;
1520         struct proc_inode *ei;
1521         const struct cred *cred;
1522
1523         /* We need a new inode */
1524
1525         inode = new_inode(sb);
1526         if (!inode)
1527                 goto out;
1528
1529         /* Common stuff */
1530         ei = PROC_I(inode);
1531         inode->i_ino = get_next_ino();
1532         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1533         inode->i_op = &proc_def_inode_operations;
1534
1535         /*
1536          * grab the reference to task.
1537          */
1538         ei->pid = get_task_pid(task, PIDTYPE_PID);
1539         if (!ei->pid)
1540                 goto out_unlock;
1541
1542         if (task_dumpable(task)) {
1543                 rcu_read_lock();
1544                 cred = __task_cred(task);
1545                 inode->i_uid = cred->euid;
1546                 inode->i_gid = cred->egid;
1547                 rcu_read_unlock();
1548         }
1549         security_task_to_inode(task, inode);
1550
1551 out:
1552         return inode;
1553
1554 out_unlock:
1555         iput(inode);
1556         return NULL;
1557 }
1558
1559 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1560 {
1561         struct inode *inode = dentry->d_inode;
1562         struct task_struct *task;
1563         const struct cred *cred;
1564         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1565
1566         generic_fillattr(inode, stat);
1567
1568         rcu_read_lock();
1569         stat->uid = GLOBAL_ROOT_UID;
1570         stat->gid = GLOBAL_ROOT_GID;
1571         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1572         if (task) {
1573                 if (!has_pid_permissions(pid, task, 2)) {
1574                         rcu_read_unlock();
1575                         /*
1576                          * This doesn't prevent learning whether PID exists,
1577                          * it only makes getattr() consistent with readdir().
1578                          */
1579                         return -ENOENT;
1580                 }
1581                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1582                     task_dumpable(task)) {
1583                         cred = __task_cred(task);
1584                         stat->uid = cred->euid;
1585                         stat->gid = cred->egid;
1586                 }
1587         }
1588         rcu_read_unlock();
1589         return 0;
1590 }
1591
1592 /* dentry stuff */
1593
1594 /*
1595  *      Exceptional case: normally we are not allowed to unhash a busy
1596  * directory. In this case, however, we can do it - no aliasing problems
1597  * due to the way we treat inodes.
1598  *
1599  * Rewrite the inode's ownerships here because the owning task may have
1600  * performed a setuid(), etc.
1601  *
1602  * Before the /proc/pid/status file was created the only way to read
1603  * the effective uid of a /process was to stat /proc/pid.  Reading
1604  * /proc/pid/status is slow enough that procps and other packages
1605  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1606  * made this apply to all per process world readable and executable
1607  * directories.
1608  */
1609 int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1610 {
1611         struct inode *inode;
1612         struct task_struct *task;
1613         const struct cred *cred;
1614
1615         if (nd && nd->flags & LOOKUP_RCU)
1616                 return -ECHILD;
1617
1618         inode = dentry->d_inode;
1619         task = get_proc_task(inode);
1620
1621         if (task) {
1622                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1623                     task_dumpable(task)) {
1624                         rcu_read_lock();
1625                         cred = __task_cred(task);
1626                         inode->i_uid = cred->euid;
1627                         inode->i_gid = cred->egid;
1628                         rcu_read_unlock();
1629                 } else {
1630                         inode->i_uid = GLOBAL_ROOT_UID;
1631                         inode->i_gid = GLOBAL_ROOT_GID;
1632                 }
1633                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1634                 security_task_to_inode(task, inode);
1635                 put_task_struct(task);
1636                 return 1;
1637         }
1638         d_drop(dentry);
1639         return 0;
1640 }
1641
1642 static int pid_delete_dentry(const struct dentry * dentry)
1643 {
1644         /* Is the task we represent dead?
1645          * If so, then don't put the dentry on the lru list,
1646          * kill it immediately.
1647          */
1648         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1649 }
1650
1651 const struct dentry_operations pid_dentry_operations =
1652 {
1653         .d_revalidate   = pid_revalidate,
1654         .d_delete       = pid_delete_dentry,
1655 };
1656
1657 /* Lookups */
1658
1659 /*
1660  * Fill a directory entry.
1661  *
1662  * If possible create the dcache entry and derive our inode number and
1663  * file type from dcache entry.
1664  *
1665  * Since all of the proc inode numbers are dynamically generated, the inode
1666  * numbers do not exist until the inode is cache.  This means creating the
1667  * the dcache entry in readdir is necessary to keep the inode numbers
1668  * reported by readdir in sync with the inode numbers reported
1669  * by stat.
1670  */
1671 int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1672         const char *name, int len,
1673         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1674 {
1675         struct dentry *child, *dir = filp->f_path.dentry;
1676         struct inode *inode;
1677         struct qstr qname;
1678         ino_t ino = 0;
1679         unsigned type = DT_UNKNOWN;
1680
1681         qname.name = name;
1682         qname.len  = len;
1683         qname.hash = full_name_hash(name, len);
1684
1685         child = d_lookup(dir, &qname);
1686         if (!child) {
1687                 struct dentry *new;
1688                 new = d_alloc(dir, &qname);
1689                 if (new) {
1690                         child = instantiate(dir->d_inode, new, task, ptr);
1691                         if (child)
1692                                 dput(new);
1693                         else
1694                                 child = new;
1695                 }
1696         }
1697         if (!child || IS_ERR(child) || !child->d_inode)
1698                 goto end_instantiate;
1699         inode = child->d_inode;
1700         if (inode) {
1701                 ino = inode->i_ino;
1702                 type = inode->i_mode >> 12;
1703         }
1704         dput(child);
1705 end_instantiate:
1706         if (!ino)
1707                 ino = find_inode_number(dir, &qname);
1708         if (!ino)
1709                 ino = 1;
1710         return filldir(dirent, name, len, filp->f_pos, ino, type);
1711 }
1712
1713 static unsigned name_to_int(struct dentry *dentry)
1714 {
1715         const char *name = dentry->d_name.name;
1716         int len = dentry->d_name.len;
1717         unsigned n = 0;
1718
1719         if (len > 1 && *name == '0')
1720                 goto out;
1721         while (len-- > 0) {
1722                 unsigned c = *name++ - '0';
1723                 if (c > 9)
1724                         goto out;
1725                 if (n >= (~0U-9)/10)
1726                         goto out;
1727                 n *= 10;
1728                 n += c;
1729         }
1730         return n;
1731 out:
1732         return ~0U;
1733 }
1734
1735 #define PROC_FDINFO_MAX 64
1736
1737 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1738 {
1739         struct task_struct *task = get_proc_task(inode);
1740         struct files_struct *files = NULL;
1741         struct file *file;
1742         int fd = proc_fd(inode);
1743
1744         if (task) {
1745                 files = get_files_struct(task);
1746                 put_task_struct(task);
1747         }
1748         if (files) {
1749                 /*
1750                  * We are not taking a ref to the file structure, so we must
1751                  * hold ->file_lock.
1752                  */
1753                 spin_lock(&files->file_lock);
1754                 file = fcheck_files(files, fd);
1755                 if (file) {
1756                         unsigned int f_flags;
1757                         struct fdtable *fdt;
1758
1759                         fdt = files_fdtable(files);
1760                         f_flags = file->f_flags & ~O_CLOEXEC;
1761                         if (close_on_exec(fd, fdt))
1762                                 f_flags |= O_CLOEXEC;
1763
1764                         if (path) {
1765                                 *path = file->f_path;
1766                                 path_get(&file->f_path);
1767                         }
1768                         if (info)
1769                                 snprintf(info, PROC_FDINFO_MAX,
1770                                          "pos:\t%lli\n"
1771                                          "flags:\t0%o\n",
1772                                          (long long) file->f_pos,
1773                                          f_flags);
1774                         spin_unlock(&files->file_lock);
1775                         put_files_struct(files);
1776                         return 0;
1777                 }
1778                 spin_unlock(&files->file_lock);
1779                 put_files_struct(files);
1780         }
1781         return -ENOENT;
1782 }
1783
1784 static int proc_fd_link(struct dentry *dentry, struct path *path)
1785 {
1786         return proc_fd_info(dentry->d_inode, path, NULL);
1787 }
1788
1789 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1790 {
1791         struct inode *inode;
1792         struct task_struct *task;
1793         int fd;
1794         struct files_struct *files;
1795         const struct cred *cred;
1796
1797         if (nd && nd->flags & LOOKUP_RCU)
1798                 return -ECHILD;
1799
1800         inode = dentry->d_inode;
1801         task = get_proc_task(inode);
1802         fd = proc_fd(inode);
1803
1804         if (task) {
1805                 files = get_files_struct(task);
1806                 if (files) {
1807                         struct file *file;
1808                         rcu_read_lock();
1809                         file = fcheck_files(files, fd);
1810                         if (file) {
1811                                 unsigned i_mode, f_mode = file->f_mode;
1812
1813                                 rcu_read_unlock();
1814                                 put_files_struct(files);
1815
1816                                 if (task_dumpable(task)) {
1817                                         rcu_read_lock();
1818                                         cred = __task_cred(task);
1819                                         inode->i_uid = cred->euid;
1820                                         inode->i_gid = cred->egid;
1821                                         rcu_read_unlock();
1822                                 } else {
1823                                         inode->i_uid = GLOBAL_ROOT_UID;
1824                                         inode->i_gid = GLOBAL_ROOT_GID;
1825                                 }
1826
1827                                 i_mode = S_IFLNK;
1828                                 if (f_mode & FMODE_READ)
1829                                         i_mode |= S_IRUSR | S_IXUSR;
1830                                 if (f_mode & FMODE_WRITE)
1831                                         i_mode |= S_IWUSR | S_IXUSR;
1832                                 inode->i_mode = i_mode;
1833
1834                                 security_task_to_inode(task, inode);
1835                                 put_task_struct(task);
1836                                 return 1;
1837                         }
1838                         rcu_read_unlock();
1839                         put_files_struct(files);
1840                 }
1841                 put_task_struct(task);
1842         }
1843         d_drop(dentry);
1844         return 0;
1845 }
1846
1847 static const struct dentry_operations tid_fd_dentry_operations =
1848 {
1849         .d_revalidate   = tid_fd_revalidate,
1850         .d_delete       = pid_delete_dentry,
1851 };
1852
1853 static struct dentry *proc_fd_instantiate(struct inode *dir,
1854         struct dentry *dentry, struct task_struct *task, const void *ptr)
1855 {
1856         unsigned fd = *(const unsigned *)ptr;
1857         struct inode *inode;
1858         struct proc_inode *ei;
1859         struct dentry *error = ERR_PTR(-ENOENT);
1860
1861         inode = proc_pid_make_inode(dir->i_sb, task);
1862         if (!inode)
1863                 goto out;
1864         ei = PROC_I(inode);
1865         ei->fd = fd;
1866
1867         inode->i_op = &proc_pid_link_inode_operations;
1868         inode->i_size = 64;
1869         ei->op.proc_get_link = proc_fd_link;
1870         d_set_d_op(dentry, &tid_fd_dentry_operations);
1871         d_add(dentry, inode);
1872         /* Close the race of the process dying before we return the dentry */
1873         if (tid_fd_revalidate(dentry, NULL))
1874                 error = NULL;
1875
1876  out:
1877         return error;
1878 }
1879
1880 static struct dentry *proc_lookupfd_common(struct inode *dir,
1881                                            struct dentry *dentry,
1882                                            instantiate_t instantiate)
1883 {
1884         struct task_struct *task = get_proc_task(dir);
1885         unsigned fd = name_to_int(dentry);
1886         struct dentry *result = ERR_PTR(-ENOENT);
1887
1888         if (!task)
1889                 goto out_no_task;
1890         if (fd == ~0U)
1891                 goto out;
1892
1893         result = instantiate(dir, dentry, task, &fd);
1894 out:
1895         put_task_struct(task);
1896 out_no_task:
1897         return result;
1898 }
1899
1900 static int proc_readfd_common(struct file * filp, void * dirent,
1901                               filldir_t filldir, instantiate_t instantiate)
1902 {
1903         struct dentry *dentry = filp->f_path.dentry;
1904         struct inode *inode = dentry->d_inode;
1905         struct task_struct *p = get_proc_task(inode);
1906         unsigned int fd, ino;
1907         int retval;
1908         struct files_struct * files;
1909
1910         retval = -ENOENT;
1911         if (!p)
1912                 goto out_no_task;
1913         retval = 0;
1914
1915         fd = filp->f_pos;
1916         switch (fd) {
1917                 case 0:
1918                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1919                                 goto out;
1920                         filp->f_pos++;
1921                 case 1:
1922                         ino = parent_ino(dentry);
1923                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1924                                 goto out;
1925                         filp->f_pos++;
1926                 default:
1927                         files = get_files_struct(p);
1928                         if (!files)
1929                                 goto out;
1930                         rcu_read_lock();
1931                         for (fd = filp->f_pos-2;
1932                              fd < files_fdtable(files)->max_fds;
1933                              fd++, filp->f_pos++) {
1934                                 char name[PROC_NUMBUF];
1935                                 int len;
1936
1937                                 if (!fcheck_files(files, fd))
1938                                         continue;
1939                                 rcu_read_unlock();
1940
1941                                 len = snprintf(name, sizeof(name), "%d", fd);
1942                                 if (proc_fill_cache(filp, dirent, filldir,
1943                                                     name, len, instantiate,
1944                                                     p, &fd) < 0) {
1945                                         rcu_read_lock();
1946                                         break;
1947                                 }
1948                                 rcu_read_lock();
1949                         }
1950                         rcu_read_unlock();
1951                         put_files_struct(files);
1952         }
1953 out:
1954         put_task_struct(p);
1955 out_no_task:
1956         return retval;
1957 }
1958
1959 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1960                                     struct nameidata *nd)
1961 {
1962         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1963 }
1964
1965 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1966 {
1967         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1968 }
1969
1970 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1971                                       size_t len, loff_t *ppos)
1972 {
1973         char tmp[PROC_FDINFO_MAX];
1974         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1975         if (!err)
1976                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1977         return err;
1978 }
1979
1980 static const struct file_operations proc_fdinfo_file_operations = {
1981         .open           = nonseekable_open,
1982         .read           = proc_fdinfo_read,
1983         .llseek         = no_llseek,
1984 };
1985
1986 static const struct file_operations proc_fd_operations = {
1987         .read           = generic_read_dir,
1988         .readdir        = proc_readfd,
1989         .llseek         = default_llseek,
1990 };
1991
1992 #ifdef CONFIG_CHECKPOINT_RESTORE
1993
1994 /*
1995  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1996  * which represent vma start and end addresses.
1997  */
1998 static int dname_to_vma_addr(struct dentry *dentry,
1999                              unsigned long *start, unsigned long *end)
2000 {
2001         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
2002                 return -EINVAL;
2003
2004         return 0;
2005 }
2006
2007 static int map_files_d_revalidate(struct dentry *dentry, struct nameidata *nd)
2008 {
2009         unsigned long vm_start, vm_end;
2010         bool exact_vma_exists = false;
2011         struct mm_struct *mm = NULL;
2012         struct task_struct *task;
2013         const struct cred *cred;
2014         struct inode *inode;
2015         int status = 0;
2016
2017         if (nd && nd->flags & LOOKUP_RCU)
2018                 return -ECHILD;
2019
2020         if (!capable(CAP_SYS_ADMIN)) {
2021                 status = -EACCES;
2022                 goto out_notask;
2023         }
2024
2025         inode = dentry->d_inode;
2026         task = get_proc_task(inode);
2027         if (!task)
2028                 goto out_notask;
2029
2030         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2031                 goto out;
2032
2033         mm = get_task_mm(task);
2034         if (!mm)
2035                 goto out;
2036
2037         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
2038                 down_read(&mm->mmap_sem);
2039                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
2040                 up_read(&mm->mmap_sem);
2041         }
2042
2043         mmput(mm);
2044
2045         if (exact_vma_exists) {
2046                 if (task_dumpable(task)) {
2047                         rcu_read_lock();
2048                         cred = __task_cred(task);
2049                         inode->i_uid = cred->euid;
2050                         inode->i_gid = cred->egid;
2051                         rcu_read_unlock();
2052                 } else {
2053                         inode->i_uid = GLOBAL_ROOT_UID;
2054                         inode->i_gid = GLOBAL_ROOT_GID;
2055                 }
2056                 security_task_to_inode(task, inode);
2057                 status = 1;
2058         }
2059
2060 out:
2061         put_task_struct(task);
2062
2063 out_notask:
2064         if (status <= 0)
2065                 d_drop(dentry);
2066
2067         return status;
2068 }
2069
2070 static const struct dentry_operations tid_map_files_dentry_operations = {
2071         .d_revalidate   = map_files_d_revalidate,
2072         .d_delete       = pid_delete_dentry,
2073 };
2074
2075 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
2076 {
2077         unsigned long vm_start, vm_end;
2078         struct vm_area_struct *vma;
2079         struct task_struct *task;
2080         struct mm_struct *mm;
2081         int rc;
2082
2083         rc = -ENOENT;
2084         task = get_proc_task(dentry->d_inode);
2085         if (!task)
2086                 goto out;
2087
2088         mm = get_task_mm(task);
2089         put_task_struct(task);
2090         if (!mm)
2091                 goto out;
2092
2093         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2094         if (rc)
2095                 goto out_mmput;
2096
2097         down_read(&mm->mmap_sem);
2098         vma = find_exact_vma(mm, vm_start, vm_end);
2099         if (vma && vma->vm_file) {
2100                 *path = vma->vm_file->f_path;
2101                 path_get(path);
2102                 rc = 0;
2103         }
2104         up_read(&mm->mmap_sem);
2105
2106 out_mmput:
2107         mmput(mm);
2108 out:
2109         return rc;
2110 }
2111
2112 struct map_files_info {
2113         struct file     *file;
2114         unsigned long   len;
2115         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
2116 };
2117
2118 static struct dentry *
2119 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2120                            struct task_struct *task, const void *ptr)
2121 {
2122         const struct file *file = ptr;
2123         struct proc_inode *ei;
2124         struct inode *inode;
2125
2126         if (!file)
2127                 return ERR_PTR(-ENOENT);
2128
2129         inode = proc_pid_make_inode(dir->i_sb, task);
2130         if (!inode)
2131                 return ERR_PTR(-ENOENT);
2132
2133         ei = PROC_I(inode);
2134         ei->op.proc_get_link = proc_map_files_get_link;
2135
2136         inode->i_op = &proc_pid_link_inode_operations;
2137         inode->i_size = 64;
2138         inode->i_mode = S_IFLNK;
2139
2140         if (file->f_mode & FMODE_READ)
2141                 inode->i_mode |= S_IRUSR;
2142         if (file->f_mode & FMODE_WRITE)
2143                 inode->i_mode |= S_IWUSR;
2144
2145         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2146         d_add(dentry, inode);
2147
2148         return NULL;
2149 }
2150
2151 static struct dentry *proc_map_files_lookup(struct inode *dir,
2152                 struct dentry *dentry, struct nameidata *nd)
2153 {
2154         unsigned long vm_start, vm_end;
2155         struct vm_area_struct *vma;
2156         struct task_struct *task;
2157         struct dentry *result;
2158         struct mm_struct *mm;
2159
2160         result = ERR_PTR(-EACCES);
2161         if (!capable(CAP_SYS_ADMIN))
2162                 goto out;
2163
2164         result = ERR_PTR(-ENOENT);
2165         task = get_proc_task(dir);
2166         if (!task)
2167                 goto out;
2168
2169         result = ERR_PTR(-EACCES);
2170         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2171                 goto out_put_task;
2172
2173         result = ERR_PTR(-ENOENT);
2174         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2175                 goto out_put_task;
2176
2177         mm = get_task_mm(task);
2178         if (!mm)
2179                 goto out_put_task;
2180
2181         down_read(&mm->mmap_sem);
2182         vma = find_exact_vma(mm, vm_start, vm_end);
2183         if (!vma)
2184                 goto out_no_vma;
2185
2186         result = proc_map_files_instantiate(dir, dentry, task, vma->vm_file);
2187
2188 out_no_vma:
2189         up_read(&mm->mmap_sem);
2190         mmput(mm);
2191 out_put_task:
2192         put_task_struct(task);
2193 out:
2194         return result;
2195 }
2196
2197 static const struct inode_operations proc_map_files_inode_operations = {
2198         .lookup         = proc_map_files_lookup,
2199         .permission     = proc_fd_permission,
2200         .setattr        = proc_setattr,
2201 };
2202
2203 static int
2204 proc_map_files_readdir(struct file *filp, void *dirent, filldir_t filldir)
2205 {
2206         struct dentry *dentry = filp->f_path.dentry;
2207         struct inode *inode = dentry->d_inode;
2208         struct vm_area_struct *vma;
2209         struct task_struct *task;
2210         struct mm_struct *mm;
2211         ino_t ino;
2212         int ret;
2213
2214         ret = -EACCES;
2215         if (!capable(CAP_SYS_ADMIN))
2216                 goto out;
2217
2218         ret = -ENOENT;
2219         task = get_proc_task(inode);
2220         if (!task)
2221                 goto out;
2222
2223         ret = -EACCES;
2224         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2225                 goto out_put_task;
2226
2227         ret = 0;
2228         switch (filp->f_pos) {
2229         case 0:
2230                 ino = inode->i_ino;
2231                 if (filldir(dirent, ".", 1, 0, ino, DT_DIR) < 0)
2232                         goto out_put_task;
2233                 filp->f_pos++;
2234         case 1:
2235                 ino = parent_ino(dentry);
2236                 if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2237                         goto out_put_task;
2238                 filp->f_pos++;
2239         default:
2240         {
2241                 unsigned long nr_files, pos, i;
2242                 struct flex_array *fa = NULL;
2243                 struct map_files_info info;
2244                 struct map_files_info *p;
2245
2246                 mm = get_task_mm(task);
2247                 if (!mm)
2248                         goto out_put_task;
2249                 down_read(&mm->mmap_sem);
2250
2251                 nr_files = 0;
2252
2253                 /*
2254                  * We need two passes here:
2255                  *
2256                  *  1) Collect vmas of mapped files with mmap_sem taken
2257                  *  2) Release mmap_sem and instantiate entries
2258                  *
2259                  * otherwise we get lockdep complained, since filldir()
2260                  * routine might require mmap_sem taken in might_fault().
2261                  */
2262
2263                 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2264                         if (vma->vm_file && ++pos > filp->f_pos)
2265                                 nr_files++;
2266                 }
2267
2268                 if (nr_files) {
2269                         fa = flex_array_alloc(sizeof(info), nr_files,
2270                                                 GFP_KERNEL);
2271                         if (!fa || flex_array_prealloc(fa, 0, nr_files,
2272                                                         GFP_KERNEL)) {
2273                                 ret = -ENOMEM;
2274                                 if (fa)
2275                                         flex_array_free(fa);
2276                                 up_read(&mm->mmap_sem);
2277                                 mmput(mm);
2278                                 goto out_put_task;
2279                         }
2280                         for (i = 0, vma = mm->mmap, pos = 2; vma;
2281                                         vma = vma->vm_next) {
2282                                 if (!vma->vm_file)
2283                                         continue;
2284                                 if (++pos <= filp->f_pos)
2285                                         continue;
2286
2287                                 get_file(vma->vm_file);
2288                                 info.file = vma->vm_file;
2289                                 info.len = snprintf(info.name,
2290                                                 sizeof(info.name), "%lx-%lx",
2291                                                 vma->vm_start, vma->vm_end);
2292                                 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2293                                         BUG();
2294                         }
2295                 }
2296                 up_read(&mm->mmap_sem);
2297
2298                 for (i = 0; i < nr_files; i++) {
2299                         p = flex_array_get(fa, i);
2300                         ret = proc_fill_cache(filp, dirent, filldir,
2301                                               p->name, p->len,
2302                                               proc_map_files_instantiate,
2303                                               task, p->file);
2304                         if (ret)
2305                                 break;
2306                         filp->f_pos++;
2307                         fput(p->file);
2308                 }
2309                 for (; i < nr_files; i++) {
2310                         /*
2311                          * In case of error don't forget
2312                          * to put rest of file refs.
2313                          */
2314                         p = flex_array_get(fa, i);
2315                         fput(p->file);
2316                 }
2317                 if (fa)
2318                         flex_array_free(fa);
2319                 mmput(mm);
2320         }
2321         }
2322
2323 out_put_task:
2324         put_task_struct(task);
2325 out:
2326         return ret;
2327 }
2328
2329 static const struct file_operations proc_map_files_operations = {
2330         .read           = generic_read_dir,
2331         .readdir        = proc_map_files_readdir,
2332         .llseek         = default_llseek,
2333 };
2334
2335 #endif /* CONFIG_CHECKPOINT_RESTORE */
2336
2337 /*
2338  * /proc/pid/fd needs a special permission handler so that a process can still
2339  * access /proc/self/fd after it has executed a setuid().
2340  */
2341 static int proc_fd_permission(struct inode *inode, int mask)
2342 {
2343         int rv = generic_permission(inode, mask);
2344         if (rv == 0)
2345                 return 0;
2346         if (task_pid(current) == proc_pid(inode))
2347                 rv = 0;
2348         return rv;
2349 }
2350
2351 /*
2352  * proc directories can do almost nothing..
2353  */
2354 static const struct inode_operations proc_fd_inode_operations = {
2355         .lookup         = proc_lookupfd,
2356         .permission     = proc_fd_permission,
2357         .setattr        = proc_setattr,
2358 };
2359
2360 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2361         struct dentry *dentry, struct task_struct *task, const void *ptr)
2362 {
2363         unsigned fd = *(unsigned *)ptr;
2364         struct inode *inode;
2365         struct proc_inode *ei;
2366         struct dentry *error = ERR_PTR(-ENOENT);
2367
2368         inode = proc_pid_make_inode(dir->i_sb, task);
2369         if (!inode)
2370                 goto out;
2371         ei = PROC_I(inode);
2372         ei->fd = fd;
2373         inode->i_mode = S_IFREG | S_IRUSR;
2374         inode->i_fop = &proc_fdinfo_file_operations;
2375         d_set_d_op(dentry, &tid_fd_dentry_operations);
2376         d_add(dentry, inode);
2377         /* Close the race of the process dying before we return the dentry */
2378         if (tid_fd_revalidate(dentry, NULL))
2379                 error = NULL;
2380
2381  out:
2382         return error;
2383 }
2384
2385 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2386                                         struct dentry *dentry,
2387                                         struct nameidata *nd)
2388 {
2389         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2390 }
2391
2392 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2393 {
2394         return proc_readfd_common(filp, dirent, filldir,
2395                                   proc_fdinfo_instantiate);
2396 }
2397
2398 static const struct file_operations proc_fdinfo_operations = {
2399         .read           = generic_read_dir,
2400         .readdir        = proc_readfdinfo,
2401         .llseek         = default_llseek,
2402 };
2403
2404 /*
2405  * proc directories can do almost nothing..
2406  */
2407 static const struct inode_operations proc_fdinfo_inode_operations = {
2408         .lookup         = proc_lookupfdinfo,
2409         .setattr        = proc_setattr,
2410 };
2411
2412
2413 static struct dentry *proc_pident_instantiate(struct inode *dir,
2414         struct dentry *dentry, struct task_struct *task, const void *ptr)
2415 {
2416         const struct pid_entry *p = ptr;
2417         struct inode *inode;
2418         struct proc_inode *ei;
2419         struct dentry *error = ERR_PTR(-ENOENT);
2420
2421         inode = proc_pid_make_inode(dir->i_sb, task);
2422         if (!inode)
2423                 goto out;
2424
2425         ei = PROC_I(inode);
2426         inode->i_mode = p->mode;
2427         if (S_ISDIR(inode->i_mode))
2428                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2429         if (p->iop)
2430                 inode->i_op = p->iop;
2431         if (p->fop)
2432                 inode->i_fop = p->fop;
2433         ei->op = p->op;
2434         d_set_d_op(dentry, &pid_dentry_operations);
2435         d_add(dentry, inode);
2436         /* Close the race of the process dying before we return the dentry */
2437         if (pid_revalidate(dentry, NULL))
2438                 error = NULL;
2439 out:
2440         return error;
2441 }
2442
2443 static struct dentry *proc_pident_lookup(struct inode *dir,
2444                                          struct dentry *dentry,
2445                                          const struct pid_entry *ents,
2446                                          unsigned int nents)
2447 {
2448         struct dentry *error;
2449         struct task_struct *task = get_proc_task(dir);
2450         const struct pid_entry *p, *last;
2451
2452         error = ERR_PTR(-ENOENT);
2453
2454         if (!task)
2455                 goto out_no_task;
2456
2457         /*
2458          * Yes, it does not scale. And it should not. Don't add
2459          * new entries into /proc/<tgid>/ without very good reasons.
2460          */
2461         last = &ents[nents - 1];
2462         for (p = ents; p <= last; p++) {
2463                 if (p->len != dentry->d_name.len)
2464                         continue;
2465                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2466                         break;
2467         }
2468         if (p > last)
2469                 goto out;
2470
2471         error = proc_pident_instantiate(dir, dentry, task, p);
2472 out:
2473         put_task_struct(task);
2474 out_no_task:
2475         return error;
2476 }
2477
2478 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2479         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2480 {
2481         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2482                                 proc_pident_instantiate, task, p);
2483 }
2484
2485 static int proc_pident_readdir(struct file *filp,
2486                 void *dirent, filldir_t filldir,
2487                 const struct pid_entry *ents, unsigned int nents)
2488 {
2489         int i;
2490         struct dentry *dentry = filp->f_path.dentry;
2491         struct inode *inode = dentry->d_inode;
2492         struct task_struct *task = get_proc_task(inode);
2493         const struct pid_entry *p, *last;
2494         ino_t ino;
2495         int ret;
2496
2497         ret = -ENOENT;
2498         if (!task)
2499                 goto out_no_task;
2500
2501         ret = 0;
2502         i = filp->f_pos;
2503         switch (i) {
2504         case 0:
2505                 ino = inode->i_ino;
2506                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2507                         goto out;
2508                 i++;
2509                 filp->f_pos++;
2510                 /* fall through */
2511         case 1:
2512                 ino = parent_ino(dentry);
2513                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2514                         goto out;
2515                 i++;
2516                 filp->f_pos++;
2517                 /* fall through */
2518         default:
2519                 i -= 2;
2520                 if (i >= nents) {
2521                         ret = 1;
2522                         goto out;
2523                 }
2524                 p = ents + i;
2525                 last = &ents[nents - 1];
2526                 while (p <= last) {
2527                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2528                                 goto out;
2529                         filp->f_pos++;
2530                         p++;
2531                 }
2532         }
2533
2534         ret = 1;
2535 out:
2536         put_task_struct(task);
2537 out_no_task:
2538         return ret;
2539 }
2540
2541 #ifdef CONFIG_SECURITY
2542 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2543                                   size_t count, loff_t *ppos)
2544 {
2545         struct inode * inode = file->f_path.dentry->d_inode;
2546         char *p = NULL;
2547         ssize_t length;
2548         struct task_struct *task = get_proc_task(inode);
2549
2550         if (!task)
2551                 return -ESRCH;
2552
2553         length = security_getprocattr(task,
2554                                       (char*)file->f_path.dentry->d_name.name,
2555                                       &p);
2556         put_task_struct(task);
2557         if (length > 0)
2558                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2559         kfree(p);
2560         return length;
2561 }
2562
2563 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2564                                    size_t count, loff_t *ppos)
2565 {
2566         struct inode * inode = file->f_path.dentry->d_inode;
2567         char *page;
2568         ssize_t length;
2569         struct task_struct *task = get_proc_task(inode);
2570
2571         length = -ESRCH;
2572         if (!task)
2573                 goto out_no_task;
2574         if (count > PAGE_SIZE)
2575                 count = PAGE_SIZE;
2576
2577         /* No partial writes. */
2578         length = -EINVAL;
2579         if (*ppos != 0)
2580                 goto out;
2581
2582         length = -ENOMEM;
2583         page = (char*)__get_free_page(GFP_TEMPORARY);
2584         if (!page)
2585                 goto out;
2586
2587         length = -EFAULT;
2588         if (copy_from_user(page, buf, count))
2589                 goto out_free;
2590
2591         /* Guard against adverse ptrace interaction */
2592         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2593         if (length < 0)
2594                 goto out_free;
2595
2596         length = security_setprocattr(task,
2597                                       (char*)file->f_path.dentry->d_name.name,
2598                                       (void*)page, count);
2599         mutex_unlock(&task->signal->cred_guard_mutex);
2600 out_free:
2601         free_page((unsigned long) page);
2602 out:
2603         put_task_struct(task);
2604 out_no_task:
2605         return length;
2606 }
2607
2608 static const struct file_operations proc_pid_attr_operations = {
2609         .read           = proc_pid_attr_read,
2610         .write          = proc_pid_attr_write,
2611         .llseek         = generic_file_llseek,
2612 };
2613
2614 static const struct pid_entry attr_dir_stuff[] = {
2615         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2616         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2617         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2618         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2619         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2620         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2621 };
2622
2623 static int proc_attr_dir_readdir(struct file * filp,
2624                              void * dirent, filldir_t filldir)
2625 {
2626         return proc_pident_readdir(filp,dirent,filldir,
2627                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2628 }
2629
2630 static const struct file_operations proc_attr_dir_operations = {
2631         .read           = generic_read_dir,
2632         .readdir        = proc_attr_dir_readdir,
2633         .llseek         = default_llseek,
2634 };
2635
2636 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2637                                 struct dentry *dentry, struct nameidata *nd)
2638 {
2639         return proc_pident_lookup(dir, dentry,
2640                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2641 }
2642
2643 static const struct inode_operations proc_attr_dir_inode_operations = {
2644         .lookup         = proc_attr_dir_lookup,
2645         .getattr        = pid_getattr,
2646         .setattr        = proc_setattr,
2647 };
2648
2649 #endif
2650
2651 #ifdef CONFIG_ELF_CORE
2652 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2653                                          size_t count, loff_t *ppos)
2654 {
2655         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2656         struct mm_struct *mm;
2657         char buffer[PROC_NUMBUF];
2658         size_t len;
2659         int ret;
2660
2661         if (!task)
2662                 return -ESRCH;
2663
2664         ret = 0;
2665         mm = get_task_mm(task);
2666         if (mm) {
2667                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2668                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2669                                 MMF_DUMP_FILTER_SHIFT));
2670                 mmput(mm);
2671                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2672         }
2673
2674         put_task_struct(task);
2675
2676         return ret;
2677 }
2678
2679 static ssize_t proc_coredump_filter_write(struct file *file,
2680                                           const char __user *buf,
2681                                           size_t count,
2682                                           loff_t *ppos)
2683 {
2684         struct task_struct *task;
2685         struct mm_struct *mm;
2686         char buffer[PROC_NUMBUF], *end;
2687         unsigned int val;
2688         int ret;
2689         int i;
2690         unsigned long mask;
2691
2692         ret = -EFAULT;
2693         memset(buffer, 0, sizeof(buffer));
2694         if (count > sizeof(buffer) - 1)
2695                 count = sizeof(buffer) - 1;
2696         if (copy_from_user(buffer, buf, count))
2697                 goto out_no_task;
2698
2699         ret = -EINVAL;
2700         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2701         if (*end == '\n')
2702                 end++;
2703         if (end - buffer == 0)
2704                 goto out_no_task;
2705
2706         ret = -ESRCH;
2707         task = get_proc_task(file->f_dentry->d_inode);
2708         if (!task)
2709                 goto out_no_task;
2710
2711         ret = end - buffer;
2712         mm = get_task_mm(task);
2713         if (!mm)
2714                 goto out_no_mm;
2715
2716         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2717                 if (val & mask)
2718                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2719                 else
2720                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2721         }
2722
2723         mmput(mm);
2724  out_no_mm:
2725         put_task_struct(task);
2726  out_no_task:
2727         return ret;
2728 }
2729
2730 static const struct file_operations proc_coredump_filter_operations = {
2731         .read           = proc_coredump_filter_read,
2732         .write          = proc_coredump_filter_write,
2733         .llseek         = generic_file_llseek,
2734 };
2735 #endif
2736
2737 /*
2738  * /proc/self:
2739  */
2740 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2741                               int buflen)
2742 {
2743         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2744         pid_t tgid = task_tgid_nr_ns(current, ns);
2745         char tmp[PROC_NUMBUF];
2746         if (!tgid)
2747                 return -ENOENT;
2748         sprintf(tmp, "%d", tgid);
2749         return vfs_readlink(dentry,buffer,buflen,tmp);
2750 }
2751
2752 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2753 {
2754         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2755         pid_t tgid = task_tgid_nr_ns(current, ns);
2756         char *name = ERR_PTR(-ENOENT);
2757         if (tgid) {
2758                 name = __getname();
2759                 if (!name)
2760                         name = ERR_PTR(-ENOMEM);
2761                 else
2762                         sprintf(name, "%d", tgid);
2763         }
2764         nd_set_link(nd, name);
2765         return NULL;
2766 }
2767
2768 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2769                                 void *cookie)
2770 {
2771         char *s = nd_get_link(nd);
2772         if (!IS_ERR(s))
2773                 __putname(s);
2774 }
2775
2776 static const struct inode_operations proc_self_inode_operations = {
2777         .readlink       = proc_self_readlink,
2778         .follow_link    = proc_self_follow_link,
2779         .put_link       = proc_self_put_link,
2780 };
2781
2782 /*
2783  * proc base
2784  *
2785  * These are the directory entries in the root directory of /proc
2786  * that properly belong to the /proc filesystem, as they describe
2787  * describe something that is process related.
2788  */
2789 static const struct pid_entry proc_base_stuff[] = {
2790         NOD("self", S_IFLNK|S_IRWXUGO,
2791                 &proc_self_inode_operations, NULL, {}),
2792 };
2793
2794 static struct dentry *proc_base_instantiate(struct inode *dir,
2795         struct dentry *dentry, struct task_struct *task, const void *ptr)
2796 {
2797         const struct pid_entry *p = ptr;
2798         struct inode *inode;
2799         struct proc_inode *ei;
2800         struct dentry *error;
2801
2802         /* Allocate the inode */
2803         error = ERR_PTR(-ENOMEM);
2804         inode = new_inode(dir->i_sb);
2805         if (!inode)
2806                 goto out;
2807
2808         /* Initialize the inode */
2809         ei = PROC_I(inode);
2810         inode->i_ino = get_next_ino();
2811         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2812
2813         /*
2814          * grab the reference to the task.
2815          */
2816         ei->pid = get_task_pid(task, PIDTYPE_PID);
2817         if (!ei->pid)
2818                 goto out_iput;
2819
2820         inode->i_mode = p->mode;
2821         if (S_ISDIR(inode->i_mode))
2822                 set_nlink(inode, 2);
2823         if (S_ISLNK(inode->i_mode))
2824                 inode->i_size = 64;
2825         if (p->iop)
2826                 inode->i_op = p->iop;
2827         if (p->fop)
2828                 inode->i_fop = p->fop;
2829         ei->op = p->op;
2830         d_add(dentry, inode);
2831         error = NULL;
2832 out:
2833         return error;
2834 out_iput:
2835         iput(inode);
2836         goto out;
2837 }
2838
2839 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2840 {
2841         struct dentry *error;
2842         struct task_struct *task = get_proc_task(dir);
2843         const struct pid_entry *p, *last;
2844
2845         error = ERR_PTR(-ENOENT);
2846
2847         if (!task)
2848                 goto out_no_task;
2849
2850         /* Lookup the directory entry */
2851         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2852         for (p = proc_base_stuff; p <= last; p++) {
2853                 if (p->len != dentry->d_name.len)
2854                         continue;
2855                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2856                         break;
2857         }
2858         if (p > last)
2859                 goto out;
2860
2861         error = proc_base_instantiate(dir, dentry, task, p);
2862
2863 out:
2864         put_task_struct(task);
2865 out_no_task:
2866         return error;
2867 }
2868
2869 static int proc_base_fill_cache(struct file *filp, void *dirent,
2870         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2871 {
2872         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2873                                 proc_base_instantiate, task, p);
2874 }
2875
2876 #ifdef CONFIG_TASK_IO_ACCOUNTING
2877 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2878 {
2879         struct task_io_accounting acct = task->ioac;
2880         unsigned long flags;
2881         int result;
2882
2883         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2884         if (result)
2885                 return result;
2886
2887         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2888                 result = -EACCES;
2889                 goto out_unlock;
2890         }
2891
2892         if (whole && lock_task_sighand(task, &flags)) {
2893                 struct task_struct *t = task;
2894
2895                 task_io_accounting_add(&acct, &task->signal->ioac);
2896                 while_each_thread(task, t)
2897                         task_io_accounting_add(&acct, &t->ioac);
2898
2899                 unlock_task_sighand(task, &flags);
2900         }
2901         result = sprintf(buffer,
2902                         "rchar: %llu\n"
2903                         "wchar: %llu\n"
2904                         "syscr: %llu\n"
2905                         "syscw: %llu\n"
2906                         "read_bytes: %llu\n"
2907                         "write_bytes: %llu\n"
2908                         "cancelled_write_bytes: %llu\n",
2909                         (unsigned long long)acct.rchar,
2910                         (unsigned long long)acct.wchar,
2911                         (unsigned long long)acct.syscr,
2912                         (unsigned long long)acct.syscw,
2913                         (unsigned long long)acct.read_bytes,
2914                         (unsigned long long)acct.write_bytes,
2915                         (unsigned long long)acct.cancelled_write_bytes);
2916 out_unlock:
2917         mutex_unlock(&task->signal->cred_guard_mutex);
2918         return result;
2919 }
2920
2921 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2922 {
2923         return do_io_accounting(task, buffer, 0);
2924 }
2925
2926 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2927 {
2928         return do_io_accounting(task, buffer, 1);
2929 }
2930 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2931
2932 #ifdef CONFIG_USER_NS
2933 static int proc_id_map_open(struct inode *inode, struct file *file,
2934         struct seq_operations *seq_ops)
2935 {
2936         struct user_namespace *ns = NULL;
2937         struct task_struct *task;
2938         struct seq_file *seq;
2939         int ret = -EINVAL;
2940
2941         task = get_proc_task(inode);
2942         if (task) {
2943                 rcu_read_lock();
2944                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2945                 rcu_read_unlock();
2946                 put_task_struct(task);
2947         }
2948         if (!ns)
2949                 goto err;
2950
2951         ret = seq_open(file, seq_ops);
2952         if (ret)
2953                 goto err_put_ns;
2954
2955         seq = file->private_data;
2956         seq->private = ns;
2957
2958         return 0;
2959 err_put_ns:
2960         put_user_ns(ns);
2961 err:
2962         return ret;
2963 }
2964
2965 static int proc_id_map_release(struct inode *inode, struct file *file)
2966 {
2967         struct seq_file *seq = file->private_data;
2968         struct user_namespace *ns = seq->private;
2969         put_user_ns(ns);
2970         return seq_release(inode, file);
2971 }
2972
2973 static int proc_uid_map_open(struct inode *inode, struct file *file)
2974 {
2975         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2976 }
2977
2978 static int proc_gid_map_open(struct inode *inode, struct file *file)
2979 {
2980         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2981 }
2982
2983 static const struct file_operations proc_uid_map_operations = {
2984         .open           = proc_uid_map_open,
2985         .write          = proc_uid_map_write,
2986         .read           = seq_read,
2987         .llseek         = seq_lseek,
2988         .release        = proc_id_map_release,
2989 };
2990
2991 static const struct file_operations proc_gid_map_operations = {
2992         .open           = proc_gid_map_open,
2993         .write          = proc_gid_map_write,
2994         .read           = seq_read,
2995         .llseek         = seq_lseek,
2996         .release        = proc_id_map_release,
2997 };
2998 #endif /* CONFIG_USER_NS */
2999
3000 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
3001                                 struct pid *pid, struct task_struct *task)
3002 {
3003         int err = lock_trace(task);
3004         if (!err) {
3005                 seq_printf(m, "%08x\n", task->personality);
3006                 unlock_trace(task);
3007         }
3008         return err;
3009 }
3010
3011 /*
3012  * Thread groups
3013  */
3014 static const struct file_operations proc_task_operations;
3015 static const struct inode_operations proc_task_inode_operations;
3016
3017 static const struct pid_entry tgid_base_stuff[] = {
3018         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
3019         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3020 #ifdef CONFIG_CHECKPOINT_RESTORE
3021         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
3022 #endif
3023         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3024         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3025 #ifdef CONFIG_NET
3026         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3027 #endif
3028         REG("environ",    S_IRUSR, proc_environ_operations),
3029         INF("auxv",       S_IRUSR, proc_pid_auxv),
3030         ONE("status",     S_IRUGO, proc_pid_status),
3031         ONE("personality", S_IRUGO, proc_pid_personality),
3032         INF("limits",     S_IRUGO, proc_pid_limits),
3033 #ifdef CONFIG_SCHED_DEBUG
3034         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3035 #endif
3036 #ifdef CONFIG_SCHED_AUTOGROUP
3037         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
3038 #endif
3039         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3040 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3041         INF("syscall",    S_IRUGO, proc_pid_syscall),
3042 #endif
3043         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
3044         ONE("stat",       S_IRUGO, proc_tgid_stat),
3045         ONE("statm",      S_IRUGO, proc_pid_statm),
3046         REG("maps",       S_IRUGO, proc_pid_maps_operations),
3047 #ifdef CONFIG_NUMA
3048         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
3049 #endif
3050         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
3051         LNK("cwd",        proc_cwd_link),
3052         LNK("root",       proc_root_link),
3053         LNK("exe",        proc_exe_link),
3054         REG("mounts",     S_IRUGO, proc_mounts_operations),
3055         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3056         REG("mountstats", S_IRUSR, proc_mountstats_operations),
3057 #ifdef CONFIG_PROC_PAGE_MONITOR
3058         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3059         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
3060         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3061 #endif
3062 #ifdef CONFIG_SECURITY
3063         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3064 #endif
3065 #ifdef CONFIG_KALLSYMS
3066         INF("wchan",      S_IRUGO, proc_pid_wchan),
3067 #endif
3068 #ifdef CONFIG_STACKTRACE
3069         ONE("stack",      S_IRUGO, proc_pid_stack),
3070 #endif
3071 #ifdef CONFIG_SCHEDSTATS
3072         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
3073 #endif
3074 #ifdef CONFIG_LATENCYTOP
3075         REG("latency",  S_IRUGO, proc_lstats_operations),
3076 #endif
3077 #ifdef CONFIG_PROC_PID_CPUSET
3078         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
3079 #endif
3080 #ifdef CONFIG_CGROUPS
3081         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3082 #endif
3083         INF("oom_score",  S_IRUGO, proc_oom_score),
3084         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3085         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3086 #ifdef CONFIG_AUDITSYSCALL
3087         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
3088         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3089 #endif
3090 #ifdef CONFIG_FAULT_INJECTION
3091         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3092 #endif
3093 #ifdef CONFIG_ELF_CORE
3094         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3095 #endif
3096 #ifdef CONFIG_TASK_IO_ACCOUNTING
3097         INF("io",       S_IRUSR, proc_tgid_io_accounting),
3098 #endif
3099 #ifdef CONFIG_HARDWALL
3100         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3101 #endif
3102 #ifdef CONFIG_USER_NS
3103         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3104         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3105 #endif
3106 };
3107
3108 static int proc_tgid_base_readdir(struct file * filp,
3109                              void * dirent, filldir_t filldir)
3110 {
3111         return proc_pident_readdir(filp,dirent,filldir,
3112                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
3113 }
3114
3115 static const struct file_operations proc_tgid_base_operations = {
3116         .read           = generic_read_dir,
3117         .readdir        = proc_tgid_base_readdir,
3118         .llseek         = default_llseek,
3119 };
3120
3121 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3122         return proc_pident_lookup(dir, dentry,
3123                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3124 }
3125
3126 static const struct inode_operations proc_tgid_base_inode_operations = {
3127         .lookup         = proc_tgid_base_lookup,
3128         .getattr        = pid_getattr,
3129         .setattr        = proc_setattr,
3130         .permission     = proc_pid_permission,
3131 };
3132
3133 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3134 {
3135         struct dentry *dentry, *leader, *dir;
3136         char buf[PROC_NUMBUF];
3137         struct qstr name;
3138
3139         name.name = buf;
3140         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3141         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3142         if (dentry) {
3143                 shrink_dcache_parent(dentry);
3144                 d_drop(dentry);
3145                 dput(dentry);
3146         }
3147
3148         name.name = buf;
3149         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
3150         leader = d_hash_and_lookup(mnt->mnt_root, &name);
3151         if (!leader)
3152                 goto out;
3153
3154         name.name = "task";
3155         name.len = strlen(name.name);
3156         dir = d_hash_and_lookup(leader, &name);
3157         if (!dir)
3158                 goto out_put_leader;
3159
3160         name.name = buf;
3161         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3162         dentry = d_hash_and_lookup(dir, &name);
3163         if (dentry) {
3164                 shrink_dcache_parent(dentry);
3165                 d_drop(dentry);
3166                 dput(dentry);
3167         }
3168
3169         dput(dir);
3170 out_put_leader:
3171         dput(leader);
3172 out:
3173         return;
3174 }
3175
3176 /**
3177  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3178  * @task: task that should be flushed.
3179  *
3180  * When flushing dentries from proc, one needs to flush them from global
3181  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3182  * in. This call is supposed to do all of this job.
3183  *
3184  * Looks in the dcache for
3185  * /proc/@pid
3186  * /proc/@tgid/task/@pid
3187  * if either directory is present flushes it and all of it'ts children
3188  * from the dcache.
3189  *
3190  * It is safe and reasonable to cache /proc entries for a task until
3191  * that task exits.  After that they just clog up the dcache with
3192  * useless entries, possibly causing useful dcache entries to be
3193  * flushed instead.  This routine is proved to flush those useless
3194  * dcache entries at process exit time.
3195  *
3196  * NOTE: This routine is just an optimization so it does not guarantee
3197  *       that no dcache entries will exist at process exit time it
3198  *       just makes it very unlikely that any will persist.
3199  */
3200
3201 void proc_flush_task(struct task_struct *task)
3202 {
3203         int i;
3204         struct pid *pid, *tgid;
3205         struct upid *upid;
3206
3207         pid = task_pid(task);
3208         tgid = task_tgid(task);
3209
3210         for (i = 0; i <= pid->level; i++) {
3211                 upid = &pid->numbers[i];
3212                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3213                                         tgid->numbers[i].nr);
3214         }
3215
3216         upid = &pid->numbers[pid->level];
3217         if (upid->nr == 1)
3218                 pid_ns_release_proc(upid->ns);
3219 }
3220
3221 static struct dentry *proc_pid_instantiate(struct inode *dir,
3222                                            struct dentry * dentry,
3223                                            struct task_struct *task, const void *ptr)
3224 {
3225         struct dentry *error = ERR_PTR(-ENOENT);
3226         struct inode *inode;
3227
3228         inode = proc_pid_make_inode(dir->i_sb, task);
3229         if (!inode)
3230                 goto out;
3231
3232         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3233         inode->i_op = &proc_tgid_base_inode_operations;
3234         inode->i_fop = &proc_tgid_base_operations;
3235         inode->i_flags|=S_IMMUTABLE;
3236
3237         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
3238                                                   ARRAY_SIZE(tgid_base_stuff)));
3239
3240         d_set_d_op(dentry, &pid_dentry_operations);
3241
3242         d_add(dentry, inode);
3243         /* Close the race of the process dying before we return the dentry */
3244         if (pid_revalidate(dentry, NULL))
3245                 error = NULL;
3246 out:
3247         return error;
3248 }
3249
3250 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3251 {
3252         struct dentry *result;
3253         struct task_struct *task;
3254         unsigned tgid;
3255         struct pid_namespace *ns;
3256
3257         result = proc_base_lookup(dir, dentry);
3258         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
3259                 goto out;
3260
3261         tgid = name_to_int(dentry);
3262         if (tgid == ~0U)
3263                 goto out;
3264
3265         ns = dentry->d_sb->s_fs_info;
3266         rcu_read_lock();
3267         task = find_task_by_pid_ns(tgid, ns);
3268         if (task)
3269                 get_task_struct(task);
3270         rcu_read_unlock();
3271         if (!task)
3272                 goto out;
3273
3274         result = proc_pid_instantiate(dir, dentry, task, NULL);
3275         put_task_struct(task);
3276 out:
3277         return result;
3278 }
3279
3280 /*
3281  * Find the first task with tgid >= tgid
3282  *
3283  */
3284 struct tgid_iter {
3285         unsigned int tgid;
3286         struct task_struct *task;
3287 };
3288 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3289 {
3290         struct pid *pid;
3291
3292         if (iter.task)
3293                 put_task_struct(iter.task);
3294         rcu_read_lock();
3295 retry:
3296         iter.task = NULL;
3297         pid = find_ge_pid(iter.tgid, ns);
3298         if (pid) {
3299                 iter.tgid = pid_nr_ns(pid, ns);
3300                 iter.task = pid_task(pid, PIDTYPE_PID);
3301                 /* What we to know is if the pid we have find is the
3302                  * pid of a thread_group_leader.  Testing for task
3303                  * being a thread_group_leader is the obvious thing
3304                  * todo but there is a window when it fails, due to
3305                  * the pid transfer logic in de_thread.
3306                  *
3307                  * So we perform the straight forward test of seeing
3308                  * if the pid we have found is the pid of a thread
3309                  * group leader, and don't worry if the task we have
3310                  * found doesn't happen to be a thread group leader.
3311                  * As we don't care in the case of readdir.
3312                  */
3313                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3314                         iter.tgid += 1;
3315                         goto retry;
3316                 }
3317                 get_task_struct(iter.task);
3318         }
3319         rcu_read_unlock();
3320         return iter;
3321 }
3322
3323 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
3324
3325 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3326         struct tgid_iter iter)
3327 {
3328         char name[PROC_NUMBUF];
3329         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
3330         return proc_fill_cache(filp, dirent, filldir, name, len,
3331                                 proc_pid_instantiate, iter.task, NULL);
3332 }
3333
3334 static int fake_filldir(void *buf, const char *name, int namelen,
3335                         loff_t offset, u64 ino, unsigned d_type)
3336 {
3337         return 0;
3338 }
3339
3340 /* for the /proc/ directory itself, after non-process stuff has been done */
3341 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
3342 {
3343         unsigned int nr;
3344         struct task_struct *reaper;
3345         struct tgid_iter iter;
3346         struct pid_namespace *ns;
3347         filldir_t __filldir;
3348
3349         if (filp->f_pos >= PID_MAX_LIMIT + TGID_OFFSET)
3350                 goto out_no_task;
3351         nr = filp->f_pos - FIRST_PROCESS_ENTRY;
3352
3353         reaper = get_proc_task(filp->f_path.dentry->d_inode);
3354         if (!reaper)
3355                 goto out_no_task;
3356
3357         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
3358                 const struct pid_entry *p = &proc_base_stuff[nr];
3359                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
3360                         goto out;
3361         }
3362
3363         ns = filp->f_dentry->d_sb->s_fs_info;
3364         iter.task = NULL;
3365         iter.tgid = filp->f_pos - TGID_OFFSET;
3366         for (iter = next_tgid(ns, iter);
3367              iter.task;
3368              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3369                 if (has_pid_permissions(ns, iter.task, 2))
3370                         __filldir = filldir;
3371                 else
3372                         __filldir = fake_filldir;
3373
3374                 filp->f_pos = iter.tgid + TGID_OFFSET;
3375                 if (proc_pid_fill_cache(filp, dirent, __filldir, iter) < 0) {
3376                         put_task_struct(iter.task);
3377                         goto out;
3378                 }
3379         }
3380         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3381 out:
3382         put_task_struct(reaper);
3383 out_no_task:
3384         return 0;
3385 }
3386
3387 /*
3388  * Tasks
3389  */
3390 static const struct pid_entry tid_base_stuff[] = {
3391         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3392         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3393         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3394         REG("environ",   S_IRUSR, proc_environ_operations),
3395         INF("auxv",      S_IRUSR, proc_pid_auxv),
3396         ONE("status",    S_IRUGO, proc_pid_status),
3397         ONE("personality", S_IRUGO, proc_pid_personality),
3398         INF("limits",    S_IRUGO, proc_pid_limits),
3399 #ifdef CONFIG_SCHED_DEBUG
3400         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3401 #endif
3402         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3403 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3404         INF("syscall",   S_IRUGO, proc_pid_syscall),
3405 #endif
3406         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
3407         ONE("stat",      S_IRUGO, proc_tid_stat),
3408         ONE("statm",     S_IRUGO, proc_pid_statm),
3409         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3410 #ifdef CONFIG_NUMA
3411         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3412 #endif
3413         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3414         LNK("cwd",       proc_cwd_link),
3415         LNK("root",      proc_root_link),
3416         LNK("exe",       proc_exe_link),
3417         REG("mounts",    S_IRUGO, proc_mounts_operations),
3418         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3419 #ifdef CONFIG_PROC_PAGE_MONITOR
3420         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3421         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3422         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3423 #endif
3424 #ifdef CONFIG_SECURITY
3425         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3426 #endif
3427 #ifdef CONFIG_KALLSYMS
3428         INF("wchan",     S_IRUGO, proc_pid_wchan),
3429 #endif
3430 #ifdef CONFIG_STACKTRACE
3431         ONE("stack",      S_IRUGO, proc_pid_stack),
3432 #endif
3433 #ifdef CONFIG_SCHEDSTATS
3434         INF("schedstat", S_IRUGO, proc_pid_schedstat),
3435 #endif
3436 #ifdef CONFIG_LATENCYTOP
3437         REG("latency",  S_IRUGO, proc_lstats_operations),
3438 #endif
3439 #ifdef CONFIG_PROC_PID_CPUSET
3440         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
3441 #endif
3442 #ifdef CONFIG_CGROUPS
3443         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3444 #endif
3445         INF("oom_score", S_IRUGO, proc_oom_score),
3446         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3447         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3448 #ifdef CONFIG_AUDITSYSCALL
3449         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3450         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3451 #endif
3452 #ifdef CONFIG_FAULT_INJECTION
3453         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3454 #endif
3455 #ifdef CONFIG_TASK_IO_ACCOUNTING
3456         INF("io",       S_IRUSR, proc_tid_io_accounting),
3457 #endif
3458 #ifdef CONFIG_HARDWALL
3459         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3460 #endif
3461 #ifdef CONFIG_USER_NS
3462         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3463         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3464 #endif
3465 };
3466
3467 static int proc_tid_base_readdir(struct file * filp,
3468                              void * dirent, filldir_t filldir)
3469 {
3470         return proc_pident_readdir(filp,dirent,filldir,
3471                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3472 }
3473
3474 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3475         return proc_pident_lookup(dir, dentry,
3476                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3477 }
3478
3479 static const struct file_operations proc_tid_base_operations = {
3480         .read           = generic_read_dir,
3481         .readdir        = proc_tid_base_readdir,
3482         .llseek         = default_llseek,
3483 };
3484
3485 static const struct inode_operations proc_tid_base_inode_operations = {
3486         .lookup         = proc_tid_base_lookup,
3487         .getattr        = pid_getattr,
3488         .setattr        = proc_setattr,
3489 };
3490
3491 static struct dentry *proc_task_instantiate(struct inode *dir,
3492         struct dentry *dentry, struct task_struct *task, const void *ptr)
3493 {
3494         struct dentry *error = ERR_PTR(-ENOENT);
3495         struct inode *inode;
3496         inode = proc_pid_make_inode(dir->i_sb, task);
3497
3498         if (!inode)
3499                 goto out;
3500         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3501         inode->i_op = &proc_tid_base_inode_operations;
3502         inode->i_fop = &proc_tid_base_operations;
3503         inode->i_flags|=S_IMMUTABLE;
3504
3505         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3506                                                   ARRAY_SIZE(tid_base_stuff)));
3507
3508         d_set_d_op(dentry, &pid_dentry_operations);
3509
3510         d_add(dentry, inode);
3511         /* Close the race of the process dying before we return the dentry */
3512         if (pid_revalidate(dentry, NULL))
3513                 error = NULL;
3514 out:
3515         return error;
3516 }
3517
3518 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3519 {
3520         struct dentry *result = ERR_PTR(-ENOENT);
3521         struct task_struct *task;
3522         struct task_struct *leader = get_proc_task(dir);
3523         unsigned tid;
3524         struct pid_namespace *ns;
3525
3526         if (!leader)
3527                 goto out_no_task;
3528
3529         tid = name_to_int(dentry);
3530         if (tid == ~0U)
3531                 goto out;
3532
3533         ns = dentry->d_sb->s_fs_info;
3534         rcu_read_lock();
3535         task = find_task_by_pid_ns(tid, ns);
3536         if (task)
3537                 get_task_struct(task);
3538         rcu_read_unlock();
3539         if (!task)
3540                 goto out;
3541         if (!same_thread_group(leader, task))
3542                 goto out_drop_task;
3543
3544         result = proc_task_instantiate(dir, dentry, task, NULL);
3545 out_drop_task:
3546         put_task_struct(task);
3547 out:
3548         put_task_struct(leader);
3549 out_no_task:
3550         return result;
3551 }
3552
3553 /*
3554  * Find the first tid of a thread group to return to user space.
3555  *
3556  * Usually this is just the thread group leader, but if the users
3557  * buffer was too small or there was a seek into the middle of the
3558  * directory we have more work todo.
3559  *
3560  * In the case of a short read we start with find_task_by_pid.
3561  *
3562  * In the case of a seek we start with the leader and walk nr
3563  * threads past it.
3564  */
3565 static struct task_struct *first_tid(struct task_struct *leader,
3566                 int tid, int nr, struct pid_namespace *ns)
3567 {
3568         struct task_struct *pos;
3569
3570         rcu_read_lock();
3571         /* Attempt to start with the pid of a thread */
3572         if (tid && (nr > 0)) {
3573                 pos = find_task_by_pid_ns(tid, ns);
3574                 if (pos && (pos->group_leader == leader))
3575                         goto found;
3576         }
3577
3578         /* If nr exceeds the number of threads there is nothing todo */
3579         pos = NULL;
3580         if (nr && nr >= get_nr_threads(leader))
3581                 goto out;
3582
3583         /* If we haven't found our starting place yet start
3584          * with the leader and walk nr threads forward.
3585          */
3586         for (pos = leader; nr > 0; --nr) {
3587                 pos = next_thread(pos);
3588                 if (pos == leader) {
3589                         pos = NULL;
3590                         goto out;
3591                 }
3592         }
3593 found:
3594         get_task_struct(pos);
3595 out:
3596         rcu_read_unlock();
3597         return pos;
3598 }
3599
3600 /*
3601  * Find the next thread in the thread list.
3602  * Return NULL if there is an error or no next thread.
3603  *
3604  * The reference to the input task_struct is released.
3605  */
3606 static struct task_struct *next_tid(struct task_struct *start)
3607 {
3608         struct task_struct *pos = NULL;
3609         rcu_read_lock();
3610         if (pid_alive(start)) {
3611                 pos = next_thread(start);
3612                 if (thread_group_leader(pos))
3613                         pos = NULL;
3614                 else
3615                         get_task_struct(pos);
3616         }
3617         rcu_read_unlock();
3618         put_task_struct(start);
3619         return pos;
3620 }
3621
3622 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3623         struct task_struct *task, int tid)
3624 {
3625         char name[PROC_NUMBUF];
3626         int len = snprintf(name, sizeof(name), "%d", tid);
3627         return proc_fill_cache(filp, dirent, filldir, name, len,
3628                                 proc_task_instantiate, task, NULL);
3629 }
3630
3631 /* for the /proc/TGID/task/ directories */
3632 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3633 {
3634         struct dentry *dentry = filp->f_path.dentry;
3635         struct inode *inode = dentry->d_inode;
3636         struct task_struct *leader = NULL;
3637         struct task_struct *task;
3638         int retval = -ENOENT;
3639         ino_t ino;
3640         int tid;
3641         struct pid_namespace *ns;
3642
3643         task = get_proc_task(inode);
3644         if (!task)
3645                 goto out_no_task;
3646         rcu_read_lock();
3647         if (pid_alive(task)) {
3648                 leader = task->group_leader;
3649                 get_task_struct(leader);
3650         }
3651         rcu_read_unlock();
3652         put_task_struct(task);
3653         if (!leader)
3654                 goto out_no_task;
3655         retval = 0;
3656
3657         switch ((unsigned long)filp->f_pos) {
3658         case 0:
3659                 ino = inode->i_ino;
3660                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3661                         goto out;
3662                 filp->f_pos++;
3663                 /* fall through */
3664         case 1:
3665                 ino = parent_ino(dentry);
3666                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3667                         goto out;
3668                 filp->f_pos++;
3669                 /* fall through */
3670         }
3671
3672         /* f_version caches the tgid value that the last readdir call couldn't
3673          * return. lseek aka telldir automagically resets f_version to 0.
3674          */
3675         ns = filp->f_dentry->d_sb->s_fs_info;
3676         tid = (int)filp->f_version;
3677         filp->f_version = 0;
3678         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3679              task;
3680              task = next_tid(task), filp->f_pos++) {
3681                 tid = task_pid_nr_ns(task, ns);
3682                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3683                         /* returning this tgid failed, save it as the first
3684                          * pid for the next readir call */
3685                         filp->f_version = (u64)tid;
3686                         put_task_struct(task);
3687                         break;
3688                 }
3689         }
3690 out:
3691         put_task_struct(leader);
3692 out_no_task:
3693         return retval;
3694 }
3695
3696 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3697 {
3698         struct inode *inode = dentry->d_inode;
3699         struct task_struct *p = get_proc_task(inode);
3700         generic_fillattr(inode, stat);
3701
3702         if (p) {
3703                 stat->nlink += get_nr_threads(p);
3704                 put_task_struct(p);
3705         }
3706
3707         return 0;
3708 }
3709
3710 static const struct inode_operations proc_task_inode_operations = {
3711         .lookup         = proc_task_lookup,
3712         .getattr        = proc_task_getattr,
3713         .setattr        = proc_setattr,
3714         .permission     = proc_pid_permission,
3715 };
3716
3717 static const struct file_operations proc_task_operations = {
3718         .read           = generic_read_dir,
3719         .readdir        = proc_task_readdir,
3720         .llseek         = default_llseek,
3721 };